We are generAI! We are looking to become the leader in combining deep learning and video creation

generAI

generai

THE AGENIÆ; THE PELOPÆUS’ VICTUALS

THE YELLOW-WINGED SPHEX

SURPRISES

EVIDENCE OF THE DESCENT OF MAN FROM SOME LOWER FORM

KEELING ISLAND: CORAL FORMATIONS

SECONDARY SEXUAL CHARACTERS OF MAMMALS—continued.

CULTIVATED PLANTS: CEREAL AND CULINARY PLANTS

THE UNFINISHED BOOK

THE LAWS OF VARIABILITY WITH RESPECT TO ANIMALS AND PLANTS

Since the emergence of generative AI, our work has undergone a significant transformation, and we need to adapt to this change. In the past, we could spend countless hours performing complex tasks to achieve a particular outcome. However, with generative AI tools, somebody on the other side can complete the same task in less time. Generative AI is not intended to replace the workforce or substitute for human intelligence. Instead, it serves as a means of enhancing our decision-making abilities by offering proposed solutions. Welcome to the era of generative AI! Amazon Q This tool is a big player in the software development industry and has been trained using billions of lines of code, using the proposed code can help us make better decisions and boost our productivity, allowing us to save time and achieve great results. This code companion supports a set of languages such as: Python
Java
Java script
Typescript
C #
and for code completion:
Go
C++
Shell
Kotlin
SQL
and for the most-supported languages in Infrastructure as Code (IaC) :
YAML (AWS cloud formation)
JSON (AWS cloud formation)
HCL (Terraform)
CDK (Typescript, Python)
Is the AI coding companion worth using for network engineers and developers? Infrastructure as code (IaC) IaC is a ubiquitous topic in the tech industry. It's mentioned in conferences, meetups, and online blogs.
thanks to the managing and provisioning of infrastructure through code instead of through manual processes, developers, and network engineers can work effectively and deliver high-quality services anytime wherever they are by collaborating easily for better results. The developer and network engineer's dilemma We all have industry-specific skills, but mastering everything in tech is difficult. Excitement and concerns surround the potential of AI.
The security concerns associated with AI arise because not all developers possess the security skills required for networking engineering or software development.
Network engineers may accidentally deploy vulnerable scripts that hackers can exploit. Developers can create apps to manage infrastructures, but if they do not follow a security-by-design approach in their software development lifecycle, their products can pose significant security concerns. Companies relying on such products may suffer reputational damage and even go bankrupt due to security breaches.
Amazon Q is a huge opportunity for network engineers and developers in the generative AI age. Where the acceleration will happen Network engineers and developers can improve their productivity by leveraging existing solutions (scripts) rather than reinventing the wheel.
the scanning provided by Q helps detect vulnerable code. Network engineers will save time by automating infrastructure deployment.
Q is an AI companion for junior engineers and developers. It accelerates their learning and helps with tasks they may find challenging.
Q has great potential for troubleshooting, as it is trained on billions of lines of code, surpassing expectations. A day with Amazon Q As a cloud engineer, I wanted to ask Q some questions for the first time as a user, here they are I asked Q what is Elastic Kubernetes Services (EKS) and here is the answer! The answer is synthetized and gives some links to the documentation, with such experience we can improve skills better and accelerate infrastructure as code with a wealth of knowledge at our fingertips. Explaining a terraform script Amazon Q explained the terraform script with details and documentation to AWS, with such experience, it’s easy to understand some scripts quickly and we can gain more time. Asking for secure coding practices The script is a snippet to provision an EC2 instance, the goal is to ask Q secure coding practices before provisioning the instance. Such experience allows developers and network engineers to improve the script before it goes into production. Conclusion In the era of generative AI, we have the opportunity to improve our learning and achieve greater results. Those who insist on reinventing the wheel cannot keep pace with teams that prioritize agility. Additionally, working with infrastructure as code is challenging and demands skilled teams to maintain high-quality infrastructures. With AI, we have an excellent opportunity to speed up infrastructure as code. We don't need to know everything; the AI will handle the rest! Since the emergence of generative AI, our work has undergone a significant transformation, and we need to adapt to this change. In the past, we could spend countless hours performing complex tasks to achieve a particular outcome. However, with generative AI tools, somebody on the other side can complete the same task in less time. Generative AI is not intended to replace the workforce or substitute for human intelligence. Instead, it serves as a means of enhancing our decision-making abilities by offering proposed solutions. Welcome to the era of generative AI! Amazon Q This tool is a big player in the software development industry and has been trained using billions of lines of code, using the proposed code can help us make better decisions and boost our productivity, allowing us to save time and achieve great results. This code companion supports a set of languages such as: Python Java Java script Typescript C # and for code completion: Go C++ Shell Kotlin SQL and for the most-supported languages in Infrastructure as Code (IaC) : YAML (AWS cloud formation) JSON (AWS cloud formation) HCL (Terraform) CDK (Typescript, Python) Is the AI coding companion worth using for network engineers and developers? Python Java Java script Typescript C # and for code completion: Go C++ Shell Kotlin SQL and for the most-supported languages in Infrastructure as Code (IaC) : YAML (AWS cloud formation) JSON (AWS cloud formation) HCL (Terraform) CDK (Typescript, Python) Is the AI coding companion worth using for network engineers and developers? Infrastructure as code (IaC) IaC is a ubiquitous topic in the tech industry. It's mentioned in conferences, meetups, and online blogs. thanks to the managing and provisioning of infrastructure through code instead of through manual processes, developers, and network engineers can work effectively and deliver high-quality services anytime wherever they are by collaborating easily for better results. IaC is a ubiquitous topic in the tech industry. It's mentioned in conferences, meetups, and online blogs. thanks to the managing and provisioning of infrastructure through code instead of through manual processes, developers, and network engineers can work effectively and deliver high-quality services anytime wherever they are by collaborating easily for better results. The developer and network engineer's dilemma We all have industry-specific skills, but mastering everything in tech is difficult. Excitement and concerns surround the potential of AI. The security concerns associated with AI arise because not all developers possess the security skills required for networking engineering or software development. Network engineers may accidentally deploy vulnerable scripts that hackers can exploit. Developers can create apps to manage infrastructures, but if they do not follow a security-by-design approach in their software development lifecycle, their products can pose significant security concerns. Companies relying on such products may suffer reputational damage and even go bankrupt due to security breaches. Amazon Q is a huge opportunity for network engineers and developers in the generative AI age. We all have industry-specific skills, but mastering everything in tech is difficult. Excitement and concerns surround the potential of AI. The security concerns associated with AI arise because not all developers possess the security skills required for networking engineering or software development. Network engineers may accidentally deploy vulnerable scripts that hackers can exploit. Developers can create apps to manage infrastructures, but if they do not follow a security-by-design approach in their software development lifecycle, their products can pose significant security concerns. Companies relying on such products may suffer reputational damage and even go bankrupt due to security breaches. Amazon Q is a huge opportunity for network engineers and developers in the generative AI age. Where the acceleration will happen Network engineers and developers can improve their productivity by leveraging existing solutions (scripts) rather than reinventing the wheel. the scanning provided by Q helps detect vulnerable code. Network engineers will save time by automating infrastructure deployment. Q is an AI companion for junior engineers and developers. It accelerates their learning and helps with tasks they may find challenging. Q has great potential for troubleshooting, as it is trained on billions of lines of code, surpassing expectations. Network engineers and developers can improve their productivity by leveraging existing solutions (scripts) rather than reinventing the wheel. the scanning provided by Q helps detect vulnerable code. Network engineers will save time by automating infrastructure deployment. Q is an AI companion for junior engineers and developers. It accelerates their learning and helps with tasks they may find challenging. Q has great potential for troubleshooting, as it is trained on billions of lines of code, surpassing expectations. A day with Amazon Q As a cloud engineer, I wanted to ask Q some questions for the first time as a user, here they are I asked Q what is Elastic Kubernetes Services (EKS) and here is the answer! The answer is synthetized and gives some links to the documentation, with such experience we can improve skills better and accelerate infrastructure as code with a wealth of knowledge at our fingertips. Explaining a terraform script Amazon Q explained the terraform script with details and documentation to AWS, with such experience, it’s easy to understand some scripts quickly and we can gain more time. Asking for secure coding practices The script is a snippet to provision an EC2 instance, the goal is to ask Q secure coding practices before provisioning the instance. Such experience allows developers and network engineers to improve the script before it goes into production. Conclusion In the era of generative AI, we have the opportunity to improve our learning and achieve greater results. Those who insist on reinventing the wheel cannot keep pace with teams that prioritize agility. Additionally, working with infrastructure as code is challenging and demands skilled teams to maintain high-quality infrastructures. With AI, we have an excellent opportunity to speed up infrastructure as code. We don't need to know everything; the AI will handle the rest!

Generative AI: Can we Accelerate Infrastructure as Code With Amazon Q?

The fight between OpenAI and Google to win the generative AI arms race has all the intensity of the war between Kendrick Lamar and Drake but without the low-blow trash talk. This week, both companies unleashed a dizzying array of updates that say a lot about their strategies: OpenAI: A More Human Gen AI On May 13, OpenAI announced several improvements to ChatGPT including multimodal capabilities. https://www.youtube.com/live/DQacCB9tDaw?si=RyXrifz6jnHbYWCW&embedable=true One feature that has captured the imagination of industry watchers is how ChatGPT is becoming more human. OpenAI demonstrated how the latest iteration of ChatGPT (GPT-4o) can detect emotions in users’ voices, analyze their facial expressions, and adjust its own tone and cadence to match user preferences. For instance, if you request a bedtime story, it can adopt a gentle, whispering tone. It can respond in a playful, sarcastic manner if you want it to. It even has the capability to sing when asked. https://youtu.be/D9byh4MAsUQ?si=DnmC0PteXYgj833j&embedable=true The demonstration immediately drew comparisons to Her, the 2013 movie in which a virtual assistant seduces an introvert… But making OpenAI more human is not in and of itself nothing new. There is a growing trend, especially among Gen Z, to turn to AI for personal companionship. Apps such as Pi offer life coaching in a helpful and conversational manner. Replika is a virtual friend and mental health lifeline. Blush is an AI-generated dating simulator. And there are many, many more examples. OpenAI knows that humanlike AI is the future, especially with Gen Z. Recently, Npj Mental Health Research studied how college students use Replika. The study observed, “Many used Replika in multiple, overlapping ways—as a friend, a therapist, and an intellectual mirror. Many also held overlapping and often conflicting beliefs about Replika—calling it a machine, an intelligence, and a human. Critically, 3% reported that Replika halted their suicidal ideation.” The significance of how ChatGPT is evolving is that OpenAI is making an incredibly popular everyday utility more human. There is no turning back now. Google: Doubling Down on AI-Powered Search At Google’s annual I/O conference on May 14, Google showed that it is leaning into its core strengths with gen AI-powered search. For instance, the AI Overview feature provides a brief, AI-generated answer to searches, along with a series of links to the sources cited in the response. Google is also introducing multistep reasoning (refining search results without requiring multiple searches in the same search session). Let’s say you are searching for a new yoga or pilates studio that is popular, conveniently located for your daily commute, and also offers a discount for new members. Soon, you will be able to ask Google Search to “find the best yoga or pilates studios in Boston and show details on their intro offers and walking time from Beacon Hill.” https://youtu.be/s4InWsd-J6g?si=X429BeoiivA6taTQ&embedable=true If you are an avid user of Google search for wayfinding, you can appreciate how important and valuable multistep reasoning will be when rolled out. The company is also introducing updates to its Gemini gen AI chatbot, including the ability to respond to longer prompts, engage in audio-based conversations, and provide more powerful functionality, including a tool summarizing 1,500 pages of the text uploaded by a user. These are important updates to a tool that I personally find more useful than ChatGPT in many ways. But for Google’s business, it is more critical that the company inject gen AI across its powerful search ecosystem, including Google Search, YouTube, and Email. Search is Google’s bread-and-butter business. And Google is fortifying its moat. Photo by Mohamed Nohassi via Unsplash The fight between OpenAI and Google to win the generative AI arms race has all the intensity of the war between Kendrick Lamar and Drake but without the low-blow trash talk. This week, both companies unleashed a dizzying array of updates that say a lot about their strategies: OpenAI: A More Human Gen AI On May 13, OpenAI announced several improvements to ChatGPT including multimodal capabilities. https://www.youtube.com/live/DQacCB9tDaw?si=RyXrifz6jnHbYWCW&embedable=true https://www.youtube.com/live/DQacCB9tDaw?si=RyXrifz6jnHbYWCW&embedable=true One feature that has captured the imagination of industry watchers is how ChatGPT is becoming more human. OpenAI demonstrated how the latest iteration of ChatGPT (GPT-4o) can detect emotions in users’ voices, analyze their facial expressions, and adjust its own tone and cadence to match user preferences. For instance, if you request a bedtime story, it can adopt a gentle, whispering tone. It can respond in a playful, sarcastic manner if you want it to. It even has the capability to sing when asked. https://youtu.be/D9byh4MAsUQ?si=DnmC0PteXYgj833j&embedable=true https://youtu.be/D9byh4MAsUQ?si=DnmC0PteXYgj833j&embedable=true The demonstration immediately drew comparisons to Her , the 2013 movie in which a virtual assistant seduces an introvert… But making OpenAI more human is not in and of itself nothing new. There is a growing trend, especially among Gen Z, to turn to AI for personal companionship. Apps such as Pi offer life coaching in a helpful and conversational manner. Replika is a virtual friend and mental health lifeline. Blush is an AI-generated dating simulator. And there are many, many more examples. The demonstration immediately drew comparisons to Her Her OpenAI knows that humanlike AI is the future, especially with Gen Z. Recently, Npj Mental Health Research studied how college students use Replika. The study observed , “Many used Replika in multiple, overlapping ways—as a friend, a therapist, and an intellectual mirror. Many also held overlapping and often conflicting beliefs about Replika—calling it a machine, an intelligence, and a human. Critically, 3% reported that Replika halted their suicidal ideation.” The study observed The significance of how ChatGPT is evolving is that OpenAI is making an incredibly popular everyday utility more human. There is no turning back now. Google: Doubling Down on AI-Powered Search At Google’s annual I/O conference on May 14, Google showed that it is leaning into its core strengths with gen AI-powered search . For instance, the AI Overview feature provides a brief, AI-generated answer to searches, along with a series of links to the sources cited in the response. Google is also introducing multistep reasoning (refining search results without requiring multiple searches in the same search session). Google showed that it is leaning into its core strengths with gen AI-powered search Let’s say you are searching for a new yoga or pilates studio that is popular, conveniently located for your daily commute, and also offers a discount for new members. Soon, you will be able to ask Google Search to “find the best yoga or pilates studios in Boston and show details on their intro offers and walking time from Beacon Hill.” https://youtu.be/s4InWsd-J6g?si=X429BeoiivA6taTQ&embedable=true https://youtu.be/s4InWsd-J6g?si=X429BeoiivA6taTQ&embedable=true If you are an avid user of Google search for wayfinding, you can appreciate how important and valuable multistep reasoning will be when rolled out. The company is also introducing updates to its Gemini gen AI chatbot, including the ability to respond to longer prompts, engage in audio-based conversations, and provide more powerful functionality, including a tool summarizing 1,500 pages of the text uploaded by a user. These are important updates to a tool that I personally find more useful than ChatGPT in many ways. But for Google’s business, it is more critical that the company inject gen AI across its powerful search ecosystem, including Google Search, YouTube, and Email. Search is Google’s bread-and-butter business. And Google is fortifying its moat. Photo by Mohamed Nohassi via Unsplash Photo by Mohamed Nohassi via Unsplash

Generative AI Clash: OpenAI’s Emotional AI vs. Google’s Enhanced Search

Hello, readers; I'm thrilled to have you join me as we explore an exciting trend in the industry known as Retrieval Augmented Generation (RAG). This cutting-edge technique is transforming how we handle customer queries, offering superior accuracy compared to traditional chatbots. With RAG, our chatbots can tap into recent data, providing more relevant responses—a capability many pre-trained Language Models (LLMs) lack. An Example to start with Let's dive into a common scenario: picture yourself managing an online retail giant like Walmart or Amazon. Your customers frequently inquire about products and other stuff. However, given the ever-changing nature of your inventory, constantly re-training a chatbot becomes impractical over time. That's where RAG heads up. By leveraging your product catalog, RAG swiftly retrieves relevant information from your data store or the knowledge base. It then generates a response with this information. This approach not only reduces training costs but also ensures seamless adaptation to dynamic data in modern businesses. Large Language Models (LLM’s) Before delving into the central idea of RAG, let's first examine a more familiar and straightforward concept: Large Language Model (LLM). Natural Language Processing (NLP) specialized models, such as LLMs are excelling at generating text by leveraging the extensive data they've been exposed to during training. These models, including ChatGPT, utilize transformer architectures to grasp the context and meaning of the input data. A famous example of a chatbot using an LLM model is the ChatGPT3.5 and GPT-4 models. In terms of LLMs, we encounter two main issues. One is the scarcity of knowledge about a specific domain that was not presented during training of LLM. The second is the generation of inaccurate and unreliable responses, often referred to as “hallucinations” resulting from insufficient exposure to recent data. To tackle these problem, an alternative approach is to fine-tune the LLM on domain-specific information. However, we will now explore a more straightforward solution involving Retrieval Augmented Generation (RAG). Retrieval Augmented Generation (RAG) Retrieval Augmented Generation (RAG) is an NLP technique that combines retrieval and generative AI models. It leverages the strengths of both retrieval models, which are good at retrieving accurate information from various data sources, and generative models, which excel at producing unique and human-like responses. By combining these capabilities, RAG aims to provide accurate and human-like responses while overcoming the limitations of each model type.  In a RAG-based AI system, a retrieval model is used to find relevant information from existing data sources while the generative model takes the retrieved information, synthesizes all the data, and shapes it into a coherent and contextually appropriate response. Components of a RAG system The primary components of the RAG include: Indexing documents: In this step, data from various sources are collected, preprocessed, and broken into chunks before being indexed into a vector database. We have previously discussed vector databases in a previous blog.
Retrieval: Once the documents are indexed, during the inference time, accurate documents are retrieved from the vector database using the similarity score between the input query and document embeddings.
Generation: After retrieving the accurate data, it is passed to a Generative AI to generate a unique and human-like response. The RAG process utilises the advantages of the other two models to deliver a more accurate answer. These valuable resources can help construct a Retrieval Augmented Generation model. Vector databases like LangChain FAISS, ChromaDB, Weaviate DB, Pinecone, Qdrant DB, and Milvus; and generative models such as GPT, Llama2, HuggingFace, an open-source platform with numerous pre-trained models for NLP, computer vision, and more. How RAG is being used today? Retrieval Augmented Generation can enhance business customer support by creating advanced chatbots that deliver swift and precise responses, resulting in increased customer satisfaction. Furthermore, this technology can generate informative knowledge base articles and help documents for your business by combining generative capabilities with domain-specific data retrieval. Resources Here are some other resources you may find helpful: IBM research blog on RAG
Medium blog with RAG implementation
Nvidia’s blog on RAG In conclusion, Retrieval Augmented Generation (RAG) marks a significant advancement in natural language processing by integrating retrieval and generative AI models. It effectively addresses the limitations of traditional chatbots and large language models, providing accurate and contextually relevant responses. Through its capability to index, retrieve, and generate information, RAG greatly enhances customer support and knowledge systems. This technology offers businesses a valuable tool for improving interactions, promising enhanced customer experiences in the digital realm. Its adaptability to dynamic data and specific queries positions RAG as a promising asset for driving business success and customer satisfaction through efficient dialogue systems. Wishing you an enjoyable and fruitful learning journey! Hello, readers; I'm thrilled to have you join me as we explore an exciting trend in the industry known as Retrieval Augmented Generation (RAG). This cutting-edge technique is transforming how we handle customer queries, offering superior accuracy compared to traditional chatbots. With RAG, our chatbots can tap into recent data, providing more relevant responses—a capability many pre-trained Language Models (LLMs) lack. An Example to start with Let's dive into a common scenario: picture yourself managing an online retail giant like Walmart or Amazon. Your customers frequently inquire about products and other stuff. However, given the ever-changing nature of your inventory, constantly re-training a chatbot becomes impractical over time. That's where RAG heads up. By leveraging your product catalog, RAG swiftly retrieves relevant information from your data store or the knowledge base. It then generates a response with this information. This approach not only reduces training costs but also ensures seamless adaptation to dynamic data in modern businesses. Let's dive into a common scenario: Large Language Models (LLM’s) Before delving into the central idea of RAG, let's first examine a more familiar and straightforward concept: Large Language Model (LLM). Natural Language Processing (NLP) specialized models, such as LLMs are excelling at generating text by leveraging the extensive data they've been exposed to during training. These models, including ChatGPT, utilize transformer architectures to grasp the context and meaning of the input data. A famous example of a chatbot using an LLM model is the ChatGPT3.5 and GPT-4 models. In terms of LLMs, we encounter two main issues. One is the scarcity of knowledge about a specific domain that was not presented during training of LLM. The second is the generation of inaccurate and unreliable responses, often referred to as “ hallucinations” resulting from insufficient exposure to recent data. To tackle these problem, an alternative approach is to fine-tune the LLM on domain-specific information. However, we will now explore a more straightforward solution involving Retrieval Augmented Generation (RAG). hallucinations” Retrieval Augmented Generation (RAG) Retrieval Augmented Generation (RAG) is an NLP technique that combines retrieval and generative AI models. It leverages the strengths of both retrieval models, which are good at retrieving accurate information from various data sources, and generative models, which excel at producing unique and human-like responses. By combining these capabilities, RAG aims to provide accurate and human-like responses while overcoming the limitations of each model type.  In a RAG-based AI system, a retrieval model is used to find relevant information from existing data sources while the generative model takes the retrieved information, synthesizes all the data, and shapes it into a coherent and contextually appropriate response. Components of a RAG system The primary components of the RAG include: Indexing documents: In this step, data from various sources are collected, preprocessed, and broken into chunks before being indexed into a vector database. We have previously discussed vector databases in a previous blog. Retrieval: Once the documents are indexed, during the inference time, accurate documents are retrieved from the vector database using the similarity score between the input query and document embeddings. Generation: After retrieving the accurate data, it is passed to a Generative AI to generate a unique and human-like response. The RAG process utilises the advantages of the other two models to deliver a more accurate answer. Indexing documents: In this step, data from various sources are collected, preprocessed, and broken into chunks before being indexed into a vector database. We have previously discussed vector databases in a previous blog . Indexing documents: previous blog Retrieval: Once the documents are indexed, during the inference time, accurate documents are retrieved from the vector database using the similarity score between the input query and document embeddings. Retrieval: Generation: After retrieving the accurate data, it is passed to a Generative AI to generate a unique and human-like response. The RAG process utilises the advantages of the other two models to deliver a more accurate answer. Generation: These valuable resources can help construct a Retrieval Augmented Generation model. Vector databases like LangChain FAISS, ChromaDB, Weaviate DB, Pinecone, Qdrant DB, and Milvus; and generative models such as GPT, Llama2, HuggingFace, an open-source platform with numerous pre-trained models for NLP, computer vision, and more. These valuable resources can help construct a Retrieval Augmented Generation model. Vector databases like LangChain FAISS, ChromaDB, Weaviate DB, Pinecone, Qdrant DB , and Milvus ; and generative models such as GPT, Llama2, HuggingFace, an open-source platform with numerous pre-trained models for NLP, computer vision , and more. LangChain FAISS, ChromaDB, Weaviate DB, Pinecone, Qdrant DB Milvus GPT, Llama2, HuggingFace, an open-source platform with numerous pre-trained models for NLP, computer vision How RAG is being used today? Retrieval Augmented Generation can enhance business customer support by creating advanced chatbots that deliver swift and precise responses, resulting in increased customer satisfaction. Furthermore, this technology can generate informative knowledge base articles and help documents for your business by combining generative capabilities with domain-specific data retrieval. Resources Here are some other resources you may find helpful: IBM research blog on RAG Medium blog with RAG implementation Nvidia’s blog on RAG IBM research blog on RAG IBM research blog on RAG Medium blog with RAG implementation Medium blog with RAG implementation Nvidia’s blog on RAG Nvidia’s blog on RAG In conclusion, Retrieval Augmented Generation (RAG) marks a significant advancement in natural language processing by integrating retrieval and generative AI models. It effectively addresses the limitations of traditional chatbots and large language models, providing accurate and contextually relevant responses. Through its capability to index, retrieve, and generate information, RAG greatly enhances customer support and knowledge systems. This technology offers businesses a valuable tool for improving interactions, promising enhanced customer experiences in the digital realm. Its adaptability to dynamic data and specific queries positions RAG as a promising asset for driving business success and customer satisfaction through efficient dialogue systems. Wishing you an enjoyable and fruitful learning journey!

Revolutionising Chatbots: The Rise of Retrieval Augmented Generation (RAG)

Inflation has brought a level of uncertainty to Wall Street in 2024 despite a record-breaking Q1 performance for the S&P 500. With much of the growth driving US markets down to the ongoing generative AI boom, could the threat of confounding consumer price index (CPI) data cause the GenAI bubble to burst? The first quarter saw the S&P 500 cross the 5,000 barrier for the first time as a watershed moment in a sustained rally that coincided with the launch of OpenAI’s ChatGPT generative AI large-language model (LLM). The index rallied more than 30% since the November 30th 2022 release of ChatGPT towards a peak of over 5,250 before the news of resurgent inflation rates forced a timely reality check. As news that CPI inflation data for March 2024 came in at a hotter-than-expected 3.5% increase in the same period in 2023, the S&P 500 tumbled back below 5,000 in a 5.46% dip from its peak in late March. Although US markets have become accustomed to high inflation data, there had been optimism that 2024 would see multiple Federal Reserve rate cuts taking place as inflation cooled. Confounding figures appeared to spark Wall Street to snap out of its long-term rally and address the prospect of higher-for-longer rates. Looking at one of the generative AI boom’s biggest stocks, Nvidia (NASDAQ:NVDA), we can see that NVDA’s momentum has slowed since March and even posted an 8% dip for the month of April. Although both the S&P 500 and Nvidia have shown resilience in the wake of the worrying inflation data, analysts remain fearful of what more bad news surrounding inflation estimates could do for undermining the momentum built by generative AI stocks. Could inflation pressures cause the generative AI bubble to finally burst? Or can we expect more resilience even in the midst of market uncertainty? Are we in a GenAI Bubble? According to Bloomberg Professional Services, the generative AI market is set to attain $1.3 trillion in revenue by 2032. However, such lofty expectations could be undermined by signs of a slowdown. “There's growing evidence that the hype machine is slowing down," notes Gerrit De Vynck, tech reporter at The Washington Post. While many major tech firms like OpenAI, Microsoft, and Google have been busy announcing new projects, the AI hype cycle is so far "yet to upend the way people work and communicate with each other." In addition to this, we’re yet to see leading firms, aside from hardware manufacturers, begin to access the massive pool of revenue forecasted for generative AI. According to John Naughton, professor of the public understanding of technology at the Open University, we’re not only in an AI bubble, but we’re already at its third-stage: euphoria. “Caution has been thrown to the winds and ostensibly rational companies are gambling colossal amounts of money on AI,” Naughton notes in his description of the euphoria stage of the AI bubble. Naughton surmises that the euphoria stage of the bubble will ultimately give way to profit-taking before reaching its conclusion at the ‘panic’ stage. Should we find ourselves inside a generative AI bubble, all eyes will be on its most impressive performer, Nvidia for signs of stress. Kristofer Barrett, equities manager at Swedbank Robur Technology, fears that the runaway success of Nvidia has pushed the firm into retail stock territory as Cathie Wood’s Ark Invest trimmed its exposure to NVDA last year while the share of ESG funds holding the stock fell to 15% at the end of 2023, down from a high of 20% in Q2 2023. Recent frailties in the S&P 500 off the back of CPI data have come at a time when investors have become accustomed to generative AI hype. For the market to be sustainable, the hype cycle will need to give way to implementation at some level. Whether or not we can reach the implementation stage of the GenAI boom will be down to whether it can deliver on its vast potential. Reasons to Remain Bullish Despite fears that the generative AI boom may be a bubble that’s readying to burst, the relative resilience of the S&P 500 which has been sustained by AI stock growth in recent months and leading GenAI stocks like Nvidia can be interpreted as a positive sign of sustainability. “Concerns about GenAI failing to meet expectations in 2024 should be seen in the context of the natural cycle of hype and maturity of new technologies,” explained Maxim Manturov, head of investment research at Freedom Finance Europe. “While some industry leaders have expressed frustration with the speed of progress in artificial intelligence and GenAI in their organizations, others remain optimistic about its potential,” Manturov added. “Companies heavily invested in GenAI are under pressure to prove profitability, and industry reports indicate a shift to more strategic AI initiatives in the coming years.” Nvidia CEO Jen Huang has corroborated the suggestion that AI initiatives are only gathering momentum in the current market. “Demand is surging worldwide across companies, industries, and nations,” Huang said. “Demand will continue to be stronger than our supply provides through the year, and we’ll do our best [to meet it].” According to Deloitte tech predictions for 2024, we’re likely to see ‘almost all’ enterprise software companies embed generative AI in at least some of their products this year. As a result, AI chip sales are set to reach more than $50 billion worldwide while software revenue could attain a $10 billion run rate by the end of 2024. Crucially, Deloitte found that more than 70% of companies are experimenting with generative AI at present, buy less than 20% are willing to spend more money than they already are on the technology. This data suggests that the pipeline of interest for GenAI solutions isn’t ready to slow down any time soon, and even if there are signs of saturation, falling entry costs would open the market for a far wider range of interested companies. If there are signs of the generative AI boom slowing, it would more likely take the form of a market correction than a bubble bursting. Could Inflation Interrupt the GenAI Boom? Tech stocks are always more vulnerable to difficult inflation data. The 2022 tech stock sell-offs on Wall Street were a timely reminder that more speculative and riskier stocks can struggle as macroeconomic conditions tighten. This means that confounding inflation data will always have an adverse impact on popular tech stocks, and given the frequency in which CPI inflation is exceeding expectations, investors may have to factor in short-term volatility when building their portfolios. For investors adopting a more long-term mindset, the prospect of generative AI flourishing into a $1.3 trillion industry will be a key draw. Whether such sky-high predictions can emerge from its hype stage into the implementation stage will be crucial, but for now, the generative AI boom is showing the resilience required for longevity. Inflation has brought a level of uncertainty to Wall Street in 2024 despite a record-breaking Q1 performance for the S&P 500. With much of the growth driving US markets down to the ongoing generative AI boom , could the threat of confounding consumer price index (CPI) data cause the GenAI bubble to burst? ongoing generative AI boom The first quarter saw the S&P 500 cross the 5,000 barrier for the first time as a watershed moment in a sustained rally that coincided with the launch of OpenAI’s ChatGPT generative AI large-language model (LLM). The index rallied more than 30% since the November 30th 2022 release of ChatGPT towards a peak of over 5,250 before the news of resurgent inflation rates forced a timely reality check. As news that CPI inflation data for March 2024 came in at a hotter-than-expected 3.5% increase in the same period in 2023, the S&P 500 tumbled back below 5,000 in a 5.46% dip from its peak in late March. hotter-than-expected 3.5% increase hotter-than-expected 3.5% increase Although US markets have become accustomed to high inflation data, there had been optimism that 2024 would see multiple Federal Reserve rate cuts taking place as inflation cooled. Confounding figures appeared to spark Wall Street to snap out of its long-term rally and address the prospect of higher-for-longer rates. Looking at one of the generative AI boom’s biggest stocks, Nvidia (NASDAQ:NVDA), we can see that NVDA’s momentum has slowed since March and even posted an 8% dip for the month of April. Although both the S&P 500 and Nvidia have shown resilience in the wake of the worrying inflation data, analysts remain fearful of what more bad news surrounding inflation estimates could do for undermining the momentum built by generative AI stocks. analysts remain fearful analysts remain fearful Could inflation pressures cause the generative AI bubble to finally burst? Or can we expect more resilience even in the midst of market uncertainty? Are we in a GenAI Bubble? According to Bloomberg Professional Services, the generative AI market is set to attain $1.3 trillion in revenue by 2032. However, such lofty expectations could be undermined by signs of a slowdown. attain $1.3 trillion in revenue attain $1.3 trillion in revenue “There's growing evidence that the hype machine is slowing down," notes Gerrit De Vynck , tech reporter at The Washington Post. While many major tech firms like OpenAI, Microsoft, and Google have been busy announcing new projects, the AI hype cycle is so far "yet to upend the way people work and communicate with each other." notes Gerrit De Vynck notes Gerrit De Vynck In addition to this, we’re yet to see leading firms, aside from hardware manufacturers, begin to access the massive pool of revenue forecasted for generative AI. According to John Naughton, professor of the public understanding of technology at the Open University, we’re not only in an AI bubble, but we’re already at its third-stage: euphoria. “Caution has been thrown to the winds and ostensibly rational companies are gambling colossal amounts of money on AI,” Naughton notes in his description of the euphoria stage of the AI bubble. Naughton notes Naughton notes Naughton surmises that the euphoria stage of the bubble will ultimately give way to profit-taking before reaching its conclusion at the ‘panic’ stage. Should we find ourselves inside a generative AI bubble, all eyes will be on its most impressive performer, Nvidia for signs of stress. Kristofer Barrett, equities manager at Swedbank Robur Technology, fears that the runaway success of Nvidia has pushed the firm into retail stock territory as Cathie Wood’s Ark Invest trimmed its exposure to NVDA last year while the share of ESG funds holding the stock fell to 15% at the end of 2023, down from a high of 20% in Q2 2023. retail stock territory retail stock territory Recent frailties in the S&P 500 off the back of CPI data have come at a time when investors have become accustomed to generative AI hype. For the market to be sustainable, the hype cycle will need to give way to implementation at some level. Whether or not we can reach the implementation stage of the GenAI boom will be down to whether it can deliver on its vast potential. Reasons to Remain Bullish Despite fears that the generative AI boom may be a bubble that’s readying to burst, the relative resilience of the S&P 500 which has been sustained by AI stock growth in recent months and leading GenAI stocks like Nvidia can be interpreted as a positive sign of sustainability. “Concerns about GenAI failing to meet expectations in 2024 should be seen in the context of the natural cycle of hype and maturity of new technologies,” explained Maxim Manturov, head of investment research at Freedom Finance Europe . Freedom Finance Europe Freedom Finance Europe “While some industry leaders have expressed frustration with the speed of progress in artificial intelligence and GenAI in their organizations, others remain optimistic about its potential,” Manturov added. “Companies heavily invested in GenAI are under pressure to prove profitability, and industry reports indicate a shift to more strategic AI initiatives in the coming years.” Nvidia CEO Jen Huang has corroborated the suggestion that AI initiatives are only gathering momentum in the current market. “Demand is surging worldwide across companies, industries, and nations,” Huang said . “Demand will continue to be stronger than our supply provides through the year, and we’ll do our best [to meet it].” Huang said Huang said According to Deloitte tech predictions for 2024 , we’re likely to see ‘almost all’ enterprise software companies embed generative AI in at least some of their products this year. As a result, AI chip sales are set to reach more than $50 billion worldwide while software revenue could attain a $10 billion run rate by the end of 2024. Deloitte tech predictions for 2024 Deloitte tech predictions for 2024 Crucially, Deloitte found that more than 70% of companies are experimenting with generative AI at present, buy less than 20% are willing to spend more money than they already are on the technology. This data suggests that the pipeline of interest for GenAI solutions isn’t ready to slow down any time soon, and even if there are signs of saturation, falling entry costs would open the market for a far wider range of interested companies. If there are signs of the generative AI boom slowing, it would more likely take the form of a market correction than a bubble bursting. Could Inflation Interrupt the GenAI Boom? Tech stocks are always more vulnerable to difficult inflation data. The 2022 tech stock sell-offs on Wall Street were a timely reminder that more speculative and riskier stocks can struggle as macroeconomic conditions tighten. This means that confounding inflation data will always have an adverse impact on popular tech stocks, and given the frequency in which CPI inflation is exceeding expectations, investors may have to factor in short-term volatility when building their portfolios. For investors adopting a more long-term mindset, the prospect of generative AI flourishing into a $1.3 trillion industry will be a key draw. Whether such sky-high predictions can emerge from its hype stage into the implementation stage will be crucial, but for now, the generative AI boom is showing the resilience required for longevity.

Could Stubborn Inflation Cause the Generative AI Bubble to Finally Burst on Wall Street?

একগুঁয়ে মুদ্রাস্ফীতি কি জেনারেটিভ এআই বুদ্বুদকে শেষ পর্যন্ত ওয়াল স্ট্রিটে ফেটে যেতে পারে?

Könnte eine hartnäckige Inflation dazu führen, dass die generative KI-Blase an der Wall Street endgültig platzt?

¿Podría la persistente inflación provocar que la burbuja generativa de la IA finalmente estalle en Wall Street?

Une inflation tenace pourrait-elle enfin faire éclater la bulle de l’IA générative à Wall Street ?

क्या जिद्दी मुद्रास्फीति के कारण वॉल स्ट्रीट पर जनरेटिव एआई बुलबुला अंततः फट सकता है?

根強いインフレにより、ウォール街で生成 AI バブルがついには崩壊する可能性があるか?

완고한 인플레이션으로 인해 생성 AI 버블이 월스트리트에서 마침내 터질 수 있습니까?

A inflação teimosa poderia fazer com que a bolha da IA generativa finalmente estourasse em Wall Street?

Может ли устойчивая инфляция привести к тому, что пузырь генеративного искусственного интеллекта наконец лопнет на Уолл-стрит?

İnatçı Enflasyon, Üretken Yapay Zeka Balonunun Sonunda Wall Street'te Patlamasına Neden Olabilir mi?

Lạm phát cứng đầu có thể khiến bong bóng AI sáng tạo cuối cùng bùng nổ trên Phố Wall?

顽固的通货膨胀会导致华尔街的生成性人工智能泡沫最终破裂吗？

Adaptive Graph Neural Networks for Cosmological Data Generalization: Acknowledgements and Disclosure

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrea Roncoli, Department of Computer, Science (University of Pisa);
(2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago);
(3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory);
(4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute);
(5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Data and Methods Results Conclusions Acknowledgments and Disclosure of Funding, and References Additional Plots 4 Conclusions We propose and demonstrate a method for unsupervised DA for cosmological inference with GNNs. We use an MMD-based loss to enable the domain-invariant encoding of features by the GNN. This approach enhances cross-domain robustness: compared to previous methods, DA-GNNs reduce prediction error and improve uncertainty estimates. Limitations The cross-domain accuracy remains worse when compared to single-domain performance. Although reaching the same accuracy might not be possible, more flexible approaches such as adversarial-based DA techniques [20, 36], instead of distance-based ones such as MMD, might yield better results. Moreover, due to computational and time constraints, our models have been trained and tested only on two of the four available CAMELS simulation suites. Using more suites would yield better cross-domain efficacy and reliability at assessment time. These limitations will be addressed in future work. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). This paper is available on arxiv under CC 4.0 license. available on arxiv Authors: Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Abstract and Intro Data and Methods Data and Methods Results Results Conclusions Conclusions Acknowledgments and Disclosure of Funding, and References Acknowledgments and Disclosure of Funding, and References Additional Plots Additional Plots 4 Conclusions We propose and demonstrate a method for unsupervised DA for cosmological inference with GNNs. We use an MMD-based loss to enable the domain-invariant encoding of features by the GNN. This approach enhances cross-domain robustness: compared to previous methods, DA-GNNs reduce prediction error and improve uncertainty estimates. Limitations The cross-domain accuracy remains worse when compared to single-domain performance. Although reaching the same accuracy might not be possible, more flexible approaches such as adversarial-based DA techniques [20, 36], instead of distance-based ones such as MMD, might yield better results. Moreover, due to computational and time constraints, our models have been trained and tested only on two of the four available CAMELS simulation suites. Using more suites would yield better cross-domain efficacy and reliability at assessment time. These limitations will be addressed in future work. Limitations

Adaptive Graph Neural Networks for Cosmological Data Generalization: Conclusions

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrea Roncoli, Department of Computer, Science (University of Pisa);
(2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago);
(3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory);
(4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute);
(5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Data and Methods Results Conclusions Acknowledgments and Disclosure of Funding, and References Additional Plots 3 Results Latent space organization Isomaps are two-dimensional projections of the multi-dimensional latent space [35]. Figure 1 shows the difference in the latent space structure without (top row) and with (bottom row) DA. Ellipses in the top right isomap highlight how the two distributions are encoded in different regions of the latent space. Without the MMD loss, the model encodes samples with very different values of Ωm close to each other, if they originate from different simulations (circles and triangles of different colors are overlapping). This scenario leads to the fragility of the regressor, which cannot learn to output different values for the same latent space encodings. On the contrary, the DA-GNN (bottom right plot) correctly encodes the samples in a domain-invariant way. Visually, circle and triangle distributions are overlapping, which indicates domain mixing. Furthermore, the direction in the color gradient shows that the DA-GNN encodes information such that the regressor can now more correctly predict cosmological parameters based on the encodings of both simulations. [5] In [41], authors get slightly better results for the same domain, and slightly worse for the cross-domain tests. We impute these differences to choices such as batch sizes and optimization techniques we took due to computational and time constraints. [6] In Appendix A, the IllustrisTNG counterpart of this plot is presented. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). This paper is available on arxiv under CC 4.0 license. available on arxiv Authors: Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Abstract and Intro Data and Methods Data and Methods Results Results Conclusions Conclusions Acknowledgments and Disclosure of Funding, and References Acknowledgments and Disclosure of Funding, and References Additional Plots Additional Plots 3 Results Latent space organization Isomaps are two-dimensional projections of the multi-dimensional latent space [35]. Figure 1 shows the difference in the latent space structure without (top row) and with (bottom row) DA. Ellipses in the top right isomap highlight how the two distributions are encoded in different regions of the latent space. Without the MMD loss, the model encodes samples with very different values of Ωm close to each other, if they originate from different simulations (circles and triangles of different colors are overlapping). This scenario leads to the fragility of the regressor, which cannot learn to output different values for the same latent space encodings. On the contrary, the DA-GNN (bottom right plot) correctly encodes the samples in a domain-invariant way. Visually, circle and triangle distributions are overlapping, which indicates domain mixing. Furthermore, the direction in the color gradient shows that the DA-GNN encodes information such that the regressor can now more correctly predict cosmological parameters based on the encodings of both simulations. [5] In [41], authors get slightly better results for the same domain, and slightly worse for the cross-domain tests. We impute these differences to choices such as batch sizes and optimization techniques we took due to computational and time constraints. [6] In Appendix A, the IllustrisTNG counterpart of this plot is presented.

Adaptive Graph Neural Networks for Cosmological Data Generalization: Results

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrea Roncoli, Department of Computer, Science (University of Pisa);
(2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago);
(3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory);
(4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute);
(5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Data and Methods Results Conclusions Acknowledgments and Disclosure of Funding, and References Additional Plots 2 Data and Methods 2.1 Domain Adaptation Optimization and Computing Resources We performed experiments on NVIDIA A100 40GB GPU. For each of the models, implemented using PyTorch Geometric [19], we perform a hyperparameter search using the Optuna library [1], with 50 trials per model. More details on code performance, model implementations, and selected hyperparameters can be found in the publicly available code[4]. 2.2 Evaluation [1] https://arepo-code.org/ [2] http://www.tapir.caltech.edu/~phopkins/Site/GIZMO.html [3] CAMELS dataset documentation: https://camels.readthedocs.io/en/latest/index.html [4] GitHub repository will be added after the anonymous review stage. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). This paper is available on arxiv under CC 4.0 license. available on arxiv Authors: Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Abstract and Intro Data and Methods Data and Methods Results Results Conclusions Conclusions Acknowledgments and Disclosure of Funding, and References Acknowledgments and Disclosure of Funding, and References Additional Plots Additional Plots 2 Data and Methods 2.1 Domain Adaptation Optimization and Computing Resources We performed experiments on NVIDIA A100 40GB GPU. For each of the models, implemented using PyTorch Geometric [19], we perform a hyperparameter search using the Optuna library [1], with 50 trials per model. More details on code performance, model implementations, and selected hyperparameters can be found in the publicly available code[4]. Optimization and Computing Resources 2.2 Evaluation [1] https://arepo-code.org/ [2] http://www.tapir.caltech.edu/~phopkins/Site/GIZMO.html [3] CAMELS dataset documentation: https://camels.readthedocs.io/en/latest/index.html [4] GitHub repository will be added after the anonymous review stage.

Adaptive Graph Neural Networks for Cosmological Data Generalization: Data and Methods

This paper is available on arxiv under CC 4.0 license.
Authors:
(1) Andrea Roncoli, Department of Computer, Science (University of Pisa);
(2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago);
(3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory);
(4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute);
(5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Data and Methods Results Conclusions Acknowledgments and Disclosure of Funding, and References Additional Plots Abstract 1 Introduction Accurate determination of cosmological parameters using big data from astronomical surveys is a task of paramount importance in modern science. Historically, the extraction of valuable cosmological information has relied on computing summary statistics [31, 16, 15]. More recently, deep learning methods, such as 2D and 3D Convolutional Neural Networks (CNNs), showed great promise in extracting rich non-linear information that summary statistics struggle to capture [32, 39, 29]. However, CNNs lack scale-invariance, as their analysis is firmly anchored to the grid size of the convolutional kernels, while any information on scales below that is lost. Choosing a superfine grid to avoid information loss, though, would simply yield almost entirely zeros in case of sparse and irregular data, such as galaxy clusterings. Thus, CNNs result in an inadequate method for structured sparse data. In contrast, Graph Neural Networks (GNNs) [23, 4, 49, 46] can handle structured cosmic web data in a scale-free manner [41, 14]. As with any other model, the typical procedure is to train GNNs on labeled data (like simulations) and then infer cosmological parameters from unlabeled data (like observations). However, there is a significant risk of these models not generalizing in the presence of the domain shift between simulations and observations. Systematic biases have been demonstrated even in experiments that train and test on simulations with different subgrid physics [41]. Domain adaptation (DA) techniques [11, 43, 18, 27] can be used to increase model robustness to this type of domain shift. Here we propose the use of Domain Adaptive Graph Neural Networks (DA-GNNs) and investigate the utility of distance-based DA losses i.e., Maximum Mean Discrepancy (MMD) [6]. MMD is an unsupervised DA technique because it does not require labeled data, which is paramount for future applications on observations. We show that our domain-adaptive models achieve stronger generalization across datasets than regular GNN models. Our work is a significant step towards building future models trained on simulations, yet robust enough to work on observational data. Related Work GNNs have shown great potential for extracting information from large sparse datasets, such as the distribution of galaxies, galaxy clusters, and cosmic large-scale structures [25, 28, 41, 33, 42, 14]. Unfortunately, due to the complexity of most deep learning models, they often learn dataset-specific features, which renders them useless when testing on a different dataset (different simulations or astronomical observations). In astronomy, it has been shown that DA techniques applied to different types of CNNs can substantially improve model performance in cross-dataset applications [8, 10, 9, 37, 21, 2]. Recently, it has been shown that DA can be used on other types of deep learning algorithms such as GNNs [12, 24, 45, 47, 7, 44, 17]. However, DA on GNNs has not been used for any astrophysics or cosmology applications. This paper is available on arxiv under CC 4.0 license. Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). This paper is available on arxiv under CC 4.0 license. available on arxiv Authors: Authors: (1) Andrea Roncoli, Department of Computer, Science (University of Pisa); (2) Aleksandra Ciprijanovi´c´, Computational Science and AI Directorate (Fermi National Accelerator Laboratory) and Department of Astronomy and Astrophysics (University of Chicago); (3) Maggie Voetberg, Computational Science and AI Directorate, (Fermi National Accelerator Laboratory); (4) Francisco Villaescusa-Navarro, Center for Computational Astrophysics (Flatiron Institute); (5) Brian Nord, Computational Science and AI Directorate, Fermi National Accelerator Laboratory, Department of Astronomy and Astrophysics (University of Chicago) and Kavli Institute for Cosmological Physics (University of Chicago). Table of Links Abstract and Intro Abstract and Intro Data and Methods Data and Methods Results Results Conclusions Conclusions Acknowledgments and Disclosure of Funding, and References Acknowledgments and Disclosure of Funding, and References Additional Plots Additional Plots Abstract 1 Introduction Accurate determination of cosmological parameters using big data from astronomical surveys is a task of paramount importance in modern science. Historically, the extraction of valuable cosmological information has relied on computing summary statistics [31, 16, 15]. More recently, deep learning methods, such as 2D and 3D Convolutional Neural Networks (CNNs), showed great promise in extracting rich non-linear information that summary statistics struggle to capture [32, 39, 29]. However, CNNs lack scale-invariance, as their analysis is firmly anchored to the grid size of the convolutional kernels, while any information on scales below that is lost. Choosing a superfine grid to avoid information loss, though, would simply yield almost entirely zeros in case of sparse and irregular data, such as galaxy clusterings. Thus, CNNs result in an inadequate method for structured sparse data. In contrast, Graph Neural Networks (GNNs) [23, 4, 49, 46] can handle structured cosmic web data in a scale-free manner [41, 14]. As with any other model, the typical procedure is to train GNNs on labeled data (like simulations) and then infer cosmological parameters from unlabeled data (like observations). However, there is a significant risk of these models not generalizing in the presence of the domain shift between simulations and observations. Systematic biases have been demonstrated even in experiments that train and test on simulations with different subgrid physics [41]. Domain adaptation (DA) techniques [11, 43, 18, 27] can be used to increase model robustness to this type of domain shift. Here we propose the use of Domain Adaptive Graph Neural Networks (DA-GNNs) and investigate the utility of distance-based DA losses i.e., Maximum Mean Discrepancy (MMD) [6]. MMD is an unsupervised DA technique because it does not require labeled data, which is paramount for future applications on observations. We show that our domain-adaptive models achieve stronger generalization across datasets than regular GNN models. Our work is a significant step towards building future models trained on simulations, yet robust enough to work on observational data. Related Work GNNs have shown great potential for extracting information from large sparse datasets, such as the distribution of galaxies, galaxy clusters, and cosmic large-scale structures [25, 28, 41, 33, 42, 14]. Unfortunately, due to the complexity of most deep learning models, they often learn dataset-specific features, which renders them useless when testing on a different dataset (different simulations or astronomical observations). In astronomy, it has been shown that DA techniques applied to different types of CNNs can substantially improve model performance in cross-dataset applications [8, 10, 9, 37, 21, 2]. Recently, it has been shown that DA can be used on other types of deep learning algorithms such as GNNs [12, 24, 45, 47, 7, 44, 17]. However, DA on GNNs has not been used for any astrophysics or cosmology applications. Related Work

Stories, News & Mentions about GENERAI

Generai’s Stories on HackerNoon