Helping 20 million+ crypto claimants worldwide unlock their trapped funds at fair market prices

OPNX (Open Exchange)

is one of the most revolutionary applications of blockchain technology. It's the
idea of turning real-world assets (RWAs) into digital representations that can live on
blockchains, opening a whole world of possibilities on how people interact and derive value
from those RWAs. Tokenization The bankruptcy claims market is, perhaps, the one space that needs this the most. An
estimated 20 million users are holding on to following the recent
wake of bankruptcies in the crypto space. And for a lot of them, tokenization of claims may
just be the respite they’ve been waiting for. claims worth $20 billion The messy business of bankruptcy claims When a company files for Chapter 11 bankruptcy, creditors can assert their right to payment
by filing a Proof of Claim with the bankruptcy court. The claim specifies the amount owed
and is usually the creditor’s only hope of getting their money back. But, having a bankruptcy claim doesn’t guarantee payment. This is because bankruptcy proceedings and restructuring financial affairs are often messy procedures. Proceedings may take years to conclude due to administrative obstacles, legal problems, and complex financial issues. And at the end of it all, not all claims are treated equally. Bankruptcy claims are assigned different priority levels. There are high-priority claims, which
must be repaid in full. Holders of other claims have to wait until these claims are sorted out.
This means lower-priority claims often don’t get paid in full or even end up totally unpaid. Thus, many creditors have grown weary of the bankruptcy process. This has led to the
development of an OTC aftermarket for bankruptcy claims. Here, creditors sell their bankruptcy claims to investors for cash, usually at a lower price than
the value of the claim. The investor hopes to make a profit down the road when the
bankruptcy proceedings conclude. Bankruptcy claims trading typically takes place with the help of bankruptcy lawyers or
agency brokers. And while it has certainly helped creditors derive some value from their
claims, it has its own share of problems. Problem with traditional claims trading Today, people who want to buy or sell bankruptcy claims can only do it via bankruptcy lawyers or brokers. However, both options can be expensive – lawyers demand high legal fees while agency brokers charge a significant portion of the transaction cost as commission. Therefore, they’re typically only accessible to larger claim holders. The market also suffers from an overall lack of transparency and information asymmetry.
Buyers don’t have convenient access to the relevant information about claims. This leaves
them vulnerable to vulture pricing. Both buyers and sellers are also often victims of payment
fraud and impersonation scams. Thus, they often have to engage the services of financial advisors or bankruptcy attorneys to
help assess the risks and opportunities of assets, which further increases the costs of claims
trading. Traditionally, the solution to all these problems was to have a whole lot of money. But this is
not something all claim holders and potential investors have. This makes it difficult for
investors to buy or sell claims, leading to another huge problem – lack of liquidity in the
bankruptcy claims market. Fortunately, blockchain technology has a better solution – tokenization. How tokenization changes things When RWAs are tokenized, they have the same financial value as the underlying assets,
with the added feature of being as tradeable as cryptocurrencies. And this is what will
revolutionize bankruptcy claims trading. Take fungibility for instance. Fungibility is the ability to exchange one asset with another of the same type without affecting its value. With tokenization, a bankruptcy claim can be converted into many digital tokens, each representing a fraction of the claim’s value. These tokens can then be freely traded on a crypto exchange with other tokens or even fiat currency. Do that for multiple bankruptcy claims and you end up with a massive pool of fungible
tokens. These can be easily traded without needing lawyers, investment advisors, or agency
brokers. This leads to a greatly reduced cost of investment. Tokenization also greatly increases the level of transparency. When claims are pooled and
made tradeable on an exchange, there will be more price transparency, leading to better
pricing. All these features make bankruptcy trading a more attractive and accessible option. It opens
up the market to both small and large claim traders. And, with more investors comes more
liquidity, which, in turn, makes trades cheaper and more accessible. Is tokenizing bankruptcy claims the future? With billions of dollars locked up in bankruptcy claims and no relief in sight, tokenization may
just be what the crypto market needs. Fortunately, this concept isn’t just theoretical
speculation. There’s a crypto company looking to tokenize claims trading. Called Open Exchange
(OPNX), the company has built an order book exchange that lets people buy and sell
tokenized claims (as well as crypto), starting with Celsius and soon, FTX. The company has standardized claims into tokens that can then be freely traded or used as collateral to trade
crypto futures. More similar solutions will likely emerge to tokenize RWAs, like claims. This means that the
next couple of years will be an interesting time in the bankruptcy claims market as
tokenization opens new avenues for trading for more people than ever before.

Tokenization: the Next Logical Step in Bankruptcy Claims Trading

The Python programming language is assisting the developers in creating standalone PC games, mobiles, and other similar enterprise applications. Python has in excess of 1, 37,000 libraries which help in many ways. In this data-centric world, most consumers demand relevant information during their buying process. The companies also need data scientists for achieving deep insights by processing the big data. This info will guide the data scientists while making critical decisions regarding streamlining business operations and several other related tasks that need valuable information for accomplishment efficiently. Therefore, with the rise in demand for data scientists, beginners and pros are looking to reach resources for learning this art of analysis and representation of data. There are some certifications programs available online which can be helpful for training. You can find blogs, videos, and other resources online as well. Let’s have a look at some of the Python Data science libraries that are helpful for you. 1 . NumPy: is among the first choice for data scientists and developers who know their technologies dealing with data-related things. This is a Python package and is available for performing scientific computations.  By using NumPy, you may leverage the n-dimensional array objects, C, C++, FORTRAN programs based on integration tools, functions for difficult mathematical operations such as Fourier transformations, linear algebra, and random numbers. Therefore you may effectively integrate the DB by selecting a variety of operations for performing. NumPy NumPy gets installed under TensorFlow and other such machine learning platforms, thereby internally providing strength to their operations. As this is an array interface, it will allow multiple options for reshaping large data sets. NumPy may be used for treating images, sound wave representations, and other binary operations. In case you have just arrived in the field of data science and machine learning, you must acquire a good understanding of NumPy for processing the real-world data sets. 2. Theano: Another useful Python library is , which assists data scientists to create big multi-dimensional arrays which are related to computing operations. This is similar to TensorFlow; however, the only difference being it is not very efficient. It involves getting used to parallel and distributed computing-related tasks. By using this, you may optimize, evaluate, or express the data-enabled mathematical operations. Theano Due to its GPU based infrastructure, the library has the capability of processing the operations in quicker ways than compared to the CPU. The library stands fit for stability and speed optimization and delivering you the expected outcome. For quicker evaluation, the C code generator used is dynamic and is extremely popular among data scientists. They can do unit testing here for identifying the flaws in the model. Read 7 Best Data Visualization Tools 3. Keras: One of the most powerful Python libraries is that permits higher-level neural network APIs for integration. The APIs will execute over the top of TensorFlow, CNTK, and Theano. Keras was developed for decreasing the challenges faced in difficult researches permitting them to compute quicker. For someone using the deep learning libraries for their work, Keras will be their best option. Keras permits quicker prototyping and supports recurrent and convoluted networks independently. It also allows various blends and execution over CPU and GPU. Keras Keras give you a user-friendly environment, thereby decreasing the efforts required for cognitive loads by using simple APIs and so providing necessary results. Because of the modular nature of Keras, you may use a range of modules from optimizers, neural layers, and activation functions, etc. for preparing newer models. Keras is an open source library and is written in Python. It is a particularly good option for the data scientists who are having trouble in adding newer models as they may easily add newer modules as functions and classes. 4. PyTorch: It is one of the largest machine learning libraries available for data scientists and researchers. The library aids them with dynamic computational graph designs; quick tensor computation accelerated via GPU and other complicated tasks. In the case of neural network algorithms, the PyTorch APIs will play an effective role. This crossbreed front-end platform is simple to use and allows transitioning into a graphical mode for optimization. In order to get precise results in the asynchronous collective operations and for the establishment of peer to peer communication, the library gives native support to its users. By using ONNX (Open Neural Network Exchange), you may export models for leveraging the visualizers, run times, platforms, and many other resources. The greatest part of PyTorch is that it enables a cloud-based environment for simple scaling of resources utilized for deployment testing. PyTorch is developed on a similar concept to another machine learning library called Torch. During the last few years, Python has gradually become more popular with the data scientists because of the trending data-centric demands. 5. SciPy: This is a Python data science library used by researchers, data scientists, and developers alike. However, do not confuse the SciPy stack with the library. SciPy gives you optimizations, integration, statistics, and linear algebra packages for the computations. The SciPy is based on the NumPy concept for dealing with difficult mathematical problems. SciPy gives numerical routines that can be used for integration and optimization. SciPy will inherit a range of sub-modules to select from. In the event that you have recently started your career in data science, SciPy will be quite helpful for guiding you through the whole numerical computation. We have seen thus far how the Python programming can assist the data scientists in analyzing and crunching big and unstructured data sets. There are other libraries such as Scikit-Learn, TensorFlow, and Eli5 available for assistance through this journey. 6 : . Pandas The Python Data Analysis Library is called PANDAS. It is an open source library in Python for availing the analysis tools and high-performance data structures. PANDAS is developed on the NumPy package, and its main data structure is DataFrame. By using DataFrame, you can manage and store data from the tables by doing manipulation of rows and columns. Methods such as square bracket notation decrease the personal effort involved in data analysis tasks such as square bracket notation. In this case, you will have the tools for accessing the data in the memory data structures and perform read and write tasks even though they are in multiple formats like SQL, CSV, Excel, or HDFS, etc. 7. PyBrain: This is a powerful modular machine learning library which is available in Python. The long form of PyBrain goes like Python Based Reinforcement Learning ArtifiArtificial Intelligence and Neural Network Library. For the entry-level data scientists, this offers flexible algorithms and modules for advanced research. It has a range of algorithms available for evolution, supervised and unsupervised learning, and neural network. For the real-life tasks PyBrain has emerged as a great tool, and it is developed across a neural network in the kernel. 8. SciKit-Learn: This is a simple tool used for data analysis and data mining-related tasks. It is licensed under BSD and is an open-source tool. It can be reused or accessed by anyone in different contexts. The SciKit is developed over NumPy, Matplottlib, and SciPy. The tool is utilized for regression, classification, and clustering or managing spam, image recognition, stock pricing, drug response, and customer segmentation, etc. SciKit-Learn allows for dimensionality reduction, pre-processing, and model selection. 9. Matplotlib: This library of Python is used for 2D plotting and is quite popular among data scientists for designing different figures in multiple formats across the respective platforms. It can be easily used in the Python code, Jupyter notebook, or IPython shells application servers. By using the Matplotlib, you will be able to make histograms, bar charts, plots, and scatter plots, etc. 10. TensorFlow: TensorFlow is an open source library designed by Google for computing the data low graphs by using empowered ML algorithms. The library was designated for fulfill fulfilling the high demands for training for neural network work. TensorFlow is not only limited to scientific computations conducted by a Google rater. It is used extensively for the popular real-world applications. Because of the flexible and high-performance architecture, you can easily deploy it for all GPUs, CPUs, or TPUs and you can perform the PC server clustering for the edge devices. 1 1. Seaborn: It was designed for visualizing complex statistical models. Seaborn comes with the potential of delivering accurate graphs like heat maps. Seaborn is developed on the Matplotlib concept, and as a result, it is highly dependent on it. Even the minor data distributions can be seen by using this library, which is the reason why the library has become popular with the developers and data scientists. 12. Bokeh: It is one of the more visualization purpose libraries used for the design of interactive plots. Similar to the library described above, this one is also developed on Matplotlib. Because of the support of used data-driven components such as D3.js this library can present interactive designs in your web browser. 13. Plotly: Now, let’s see the description of Plotly, which happens to be one of the most popular web-based frameworks used by the data scientists. The toolbox offers the design of visualization models by using a range of API varieties supported by the multiple programming languages which include Python. InterInteractive graphics can be easily used along with numerous robust accessories via the main site plot.ly. For utilizing Plotly in the working model, you will have to set up the available API keys correctly. The graphics are processed on the server side, and once they are successfully executed, they will start appearing on the browser screen. 14. NLTK: The long form of NLTK is Natural Language ToolKit. As indicated by its name, the library is useful in accomplishing natural language processing tasks. In the beginning, it was created for promoting teaching models along with other NLP enabled research like the cognitive theory used in AI and linguistic models. It has been a successful resource in its area and drives real-world innovations of artificial intelligence. By using NLTK you can perform operations such as stemming, text tagging, regression, corpus tree creation, semantic reasoning, named entities recognition, tokenization, classifications and a range of other difficult AI related tasks. Now challenging work will need large building blocks such as semantic analysis, summarization, and automation. But this work has become easier and can be easily accomplished by using NLTK. : 15. Gensim It is a Python-based open-source library that permits topic modeling and space vector computation by using an implemented range of tools. compatible with the big test and makes for efficient operation and in-memory processing. It utilizes SciPy and NumPy modules to provide easy and efficient handling of the environment. Gensim utilizes unstructured digital text and processes it by using in-built algorithms such as word2vec, Latent Dirichlet Allocation (LDA), Hierarchical Dirichlet Processes (HDP), and Latent Semantic Analysis (LSA). 16. Scrapy: Scrapy is also known as spider bots. Scrapy is a Data science library responsible for crawling the programs and retrieving structured data out of web applications. Scrapy is a Python-written open-source library. This happens to be a complete framework having the potential to collect data via APIs and acts as a crawler. You can write codes by using Scrapy, re-utilize universal programs, and develop scalable crawlers for the applications. it is created across a spider class that contains instructions for the crawler. 17. Statsmodels: Statsmodels is another Python library, and it is responsible for giving exploration modules by using multiple methods for performing assertions and statistical analysis. It uses robust linear models, time series, analysis models, regression techniques, and discrete choice models, thereby making it prominent among similar data science libraries. It comes with a plotting function for the statistical analysis for achieving high-performance outcomes during the processing of the large statistical data sets. 18. Kivy: This is another open-source Python library providing a natural user interface that may be accessed easily over Linux, Windows, or Android. The open-source library is licensed under MIT, and it is quite helpful in the building of mobile apps along with multi-touch applications. In the beginning, the library was developed for the Kivy iOS and came with features such as a graphics library. Extensive support is provided to the hardware with a keyboard, mouse, and a range of widgets. You can also use Kivy for creating custom widgets by applying it as an intermediate language. 19. PyQt: Another Python binding toolkit for being used as a cross-platform GUI is PyQt. PyQt is being implemented as the Python plugin. It is a free application licensed under the General Public License (GNU). It comes with around 440 classes and in excess of 6000 functions in order to make the user experience simpler. PyQt has classes to access SQL databases, active X controller classes, an XML parser, SVG support, and several other useful resources for reducing user challenges. 20. OpenCV: This library is designed for driving the growth of real-time computation application development. The library is created by Intel, and the open-source platform is licensed with BSD. It is free for use by anyone. OpenCV comes with 2D and 3D feature toolkits, mobile robotics, gesture recognition, SFM, Naive Bayes classifier, gradient boosting trees, AR boosting, motion tracking, segmentation, face recognition and object identification algorithms. Although OpenCV is written by using C++, it will provide binding with Python, Octave, and Java. Previously published at https://skillpaythebills.com/top-20-data-science-libraries/

Python Libraries For Data Science

Object Detection Frameworks That Will Dominate 2023 and Beyond

Shape of the near future — multi-assistant world, data slavery, justice and drones — depends on the resolution of some hard choices. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ If Duplex is the future, then Google Assistant, Alexa, and Cortana should start working together — “[ …[is] conversational AI that makes phone calls to schedule appointments or make restaurant reservations**….** Google] Duplex …. [Since] bot-to-bot communication will soon be critical…makers of assistants like Siri, Alexa, and Google Assistant need to work together Amazon and Microsoft are planning to provide [this [because] we will come to live in a multi-assistant world ]… …. An example of how this could function is the Open Neural Network Exchange (ONNX) format made available last year for interoperability of Microsoft’s Cognitive Toolkit, Facebook’s Caffe2, and PyTorch…[with] an ecosystem of engines and frameworks to train and deploy AI …. .” [This] would also boost the entire conversational computing space…to entrust more tasks…[and] make room…for specialized assistants…[to] do their job more effectively or accomplish other goals….[A] very different world from the one we live in today https://venturebeat.com/2018/07/06/ai-weekly-if-duplex-is-the-future-then-google-assistant-alexa-and-cortana-should-start-working-together/ _Is the Content Worth the Price?_medium.com Virtual Assistants — Where’s the Beef? What if people were paid for their data? — “‘DATA SLAVERY’…is the state in which most people now live…[and] something is rotten in the data economy…. …. …. Germany’s chancellor, recently called for a price to be put on personal data [If] people really controlled their data — and the tech giants were required to pay for access…[w]hat would such a data economy look like? Unlike physical resources, personal data are…‘non-rival’ goods, meaning…the more they are used, the better for society…. [Since] most AI algorithms need to be trained using reams of human-generated examples…[it’s] a form of labor which powers AI …[mostly] passive, as people engage in all kinds of activities….[Some’s] more active, as they make decisions (such as labelling images or steering a car through a busy city …. [Still] few people will have the time or inclination to keep track of all the information they generate, or estimate its value…[and] lack the bargaining power to get a good deal from AI firms…. … [Expect] rise of…‘data-labor unions’…[to] serve as gatekeepers of people’s data [though] so far failed to make much headway…. …. [The] tech giants already pay for data…through outsourcing…armies of raters and moderators to check the quality of their algorithms and take down content that is illegal or offensive. Other firms use crowd-working platforms [ ….[But] to happen…. I]f AI lives up to the hype, it will lead to demand for more and better data…which people may only provide if they get paid [ First] legal framework will be needed to encourage the emergence of a new data economy…. Second, the technology to keep track of data flows needs to become much more capable. Research to calculate the value of particular data to an AI service is in its infancy…. .…. , [Third] people will have to develop a ‘class consciousness’ as data workers 90% of Americans think being in control of who can get data on them is important according to the Pew Research …. .” [Societies] have to find a mechanism to distribute the wealth created by AI. As things stand, most of it accrues to the big data distilleries. Unless this changes, social inequality could revert to medieval levels https://www.economist.com/the-world-if/2018/07/07/what-if-people-were-paid-for-their-data _What’s Yours is Mine and What’s Mine is Mine_medium.com My Data Is My Property — Pay Me If You Use It If the Supreme Court is Nakedly Political, Can It Be Just? — “[W]e wonder whether a Supreme Court that has come to be rigidly divided by both ideology and party can sustain public confidence for much longer….[in] era of partisan division…. …[and in] The trend is extreme — and alarming. In the 1950s and 1960s, the ideological biases…appointees were relatively modest. The gap…has steadily grown past 10 years…[and] justices have hardly ever voted against the ideology of the president who appointed them…. …[ …. [It’s] impossible to regard the court as anything but a partisan institution and] will be seen by the public as one whose votes on controversial issues are essentially determined by the party affiliation of recent presidents ” Assaults on judicial independence are…easier when the public comes to view the judiciary as a political body. This risk…should be at the center of public attention. https://www.nytimes.com/2018/07/09/opinion/supreme-court-nominee-trump.html?rref=collection%2Fsectioncollection%2Fopinion&action=click&contentCollection=opinion&region=rank&module=package&version=highlights&contentPlacement=8&pgtype=sectionfront Tomorrow’s squadron leaders will be accompanied by drones - “ is about the size of a small jet fighter…shaped like a flying wing….[is] an unmanned, stealthy combat drone…[operated] by a pilot sitting in a control center on the ground…. [ Taranis] [When] using artificially intelligent [systems]… can take off, find its way to a combat zone, select a target, attack said target with missiles and then find its way home and land…. … … …. [A] single pilot could fly with a number of combat drones…as his ‘wingbots’… might be used to reconnoiter an area or attack it, permitting the manned aircraft to hold back….[They can] undertake defensive…turns at g-forces so high that a human being could not withstand to divert incoming missiles away from a manned fighter and issuing orders to accompanying combat drones [ Many] countries are now exploring…unmanned military planes to carry cargo and as refueling tankers ….[Airbus] looking at single-pilot operation…[with] a ground-based pilot to take control of an aircraft in the event of a problem.” https://www.economist.com/science-and-technology/2018/07/05/tomorrows-squadron-leaders-will-be-accompanied-by-drones Find more of my ideas on Medium at, . A Passion to Evolve Or click the Follow button below to add me to your feed. Prefer a weekly email newsletter — free, no ads, no spam, & never sell the list — email me with “add me” in the subject line. dochuston1@gmail.com May you live long and prosper!Doc Huston

News — At The Edge — 7/14

We are big believers in community-building. We have invested heavily in it as a firm, contributing to the ecosystem in various ways that are consistent with our values (e.g., fostering , supporting and ). And we have invested heavily in it across our portfolio companies. Our insight a decade ago that developers were becoming the kings of IT led to a fruitful investment thesis, led by my partner , around and community building (as exemplified by our investments in MongoDB, Codecademy, Crashlytics and Firebase). diversity in tech women entrepreneurs immigrant entrepreneurs Chip developer driven adoption For the last few years, we have been applying this thesis to our investments in and . Blockchain-based startups share many of the same characteristics as developer-driven startups. With these companies, sustainable competitive advantage is built through a loyal community of supporters who contribute to the project and feel ownership over its success. Because of , a Blockchain startup has the opportunity to create ardent supporters that literally have a stake in the project in ways that MongoDB and other open source projects could never do. blockchain crypto token network effects On the eve of our second portfolio company completing an Initial Coin Offering, we gathered the community managers of our five Blockchain startups and — together with our advisors, MongoDB’s (first nontechnical hire and head of community and demand generation) and OpenX’s (founding CTO of the largest open source advertising exchange network) — to share best practices in community-building in the context of crypto. Here’s what we learned. Meghan Gill Scott Switzer Provide An Amazing First-Time Experience MongoDB is a complex, scalable database for sophisticated customers. But their design goal was to provide an amazing out of the box experience — holding themselves to the standard of being able to download, install and make use of the product in under five minutes. Crashlytics founder, , likes to refer to the initial product you release to your community as the Minimum Lovable Product (MLP): a product that is “so intuitive, so satisfying to use, that your customer base can’t help but tell someone about their experience.” Focus on triggering an emotional response, Chang advises, with a narrative and small surprises that compel the user to appreciate that you are thinking of them and their needs. Wayne Chang Jump On An Existing Wave One of the most important pieces of advice we heard from our community-building experts was the value of latching on to existing communities that already have momentum behind them. , a decentralized crypto exchange, has latched on to the community (a non-profit the founders started to support the NEO ecosystem). MongoDB latched on to the NoSQL movement. OpenX latched on to the rise in programmatic advertising. If you can frame your work in the context of an existing trend or platform and then establish your presence in the communities and academic environments where that trend is flourishing, you can put yourself in a strong position to ride someone else’s wave and not be forced to generate all the kinetic energy to support your platform by yourself. NEX City of Zion There’s No Substitution For Presence, Especially The Founder Megan advises, “There’s no magic bullet when it comes to community building. You have to be consistent and always available on line — wherever your community wants to interact with you.” This presence has been challenging for blockchain startups as there is already tremendous fragmentation in communications channels. Discord, Telegram, Slack, Twitter, Reddit, Stack Overflow and many others are all very active forums for different segments of your community. And, if you want to jump on an existing wave, you need to be active through all those channels in the communities of others as well. The team from (a scalable privacy protocol that enables “secret contracts”) talked about their heightened sense of urgency in being organized to answer questions that come up online immediately. As OpenX’s Switzer put it, “Being available to communicate with and respond to the community is one of the biggest jobs I have as a founder”. Enigma recently created an in order to help scale the presence of their small team. Ambassadors have their own channel and private calls with the founders to get briefed on the product roadmap and upcoming releases. Enigma built their own scraper / web crawler to track all the questions that come from the different forums and loads them automatically into their issue tracking platform, Jira. (a network-based scalability solution for blockchains) is focused on attracting a deeply technical community at this stage in their roadmap and so have put out more technical content to match what their community might be looking for. Enigma Ambassador Program BloXroute In watching our blockchain-based startups evolve, we have come to appreciate more and more the importance of a subtle skill on the part of founding teams in crypto. They not only need to have the usual compelling vision, unique market insight and execution ability. They also need to be intentional, strategic, capable community builders. Thanks to Samir Ghosh, Meghan Gill and Scott Switzer for their contributions — as well as Tor Bair (Enigma), Kevin Flynn, Carla Paiva, Giuseppe Stuto and Nissa Szabo.

Crypto Community Building

Deploying machine learning models outside of a environment used to be difficult. When the target platform is the browser, the defacto standard for serving predictions has been an API call to a server-side inference engine. For many reasons, server-side inference APIs are a non-optimal solution and machine learning models are more often being deployed natively. has done a good job at supporting this movement by providing cross-platform APIs, however many of us do not want to be married to a single ecosystem. Python TensorFlow In comes the (ONNX) project which, since being picked up by , has been seeing massive development efforts and is approaching a stable state. It's now easier than ever to deploy machine-learning models; trained using your machine-learning framework of choice, on your platform of choice, with hardware acceleration out of the box. Open Neural Network Exchange Microsoft In April this year, was introduced (see this ). uses to compile the inference engine to format - it's about time started to flex its muscles. Especially when paired with we suddenly have GPU-powered machine learning in the browser, pretty cool. onnxruntime-web Pull Request onnxruntime-web WebAssembly onnxruntime wasm WebAssembly WebGL, In this tutorial we will dive into by deploying a pre-trained model to the browser. We will be using as our deployment target. AlexNet has been trained as an image classifier on the , so we will be building an image classifier - nothing better than re-inventing the wheel. At the end of this tutorial, we will have built a bundled web app that can be run as a stand alone static web page, or integrated into your framework of choice. onnxruntime-web PyTorch AlexNet ImageNet dataset JavaScript → Jump to code onnxruntime-web-tutorial Prerequisite You will need a trained machine-learning model exported as an binary protobuf file. There's many ways to achieve this using a number of different deep-learning frameworks. For the sake of this tutorial, I will be using the exported model from the example in the , the python code snippet below will help you generate your own model. You can also follow the documentation to export your own model. If you're coming from , will help you with exporting your model to . Lastly, doesn't just pride itself on cross-platform deployment, but also in allowing exports from all major deep-learning frameworks. Those of you using another deep learning framework should be able to find support for exporting to in the docs of your framework. ONNX AlexNet PyTorch documentation PyTorch Tensorflow this tutorial ONNX ONNX ONNX import torch
import torchvision

dummy_input = torch.randn(1, 3, 224, 224)
model = torchvision.models.alexnet(pretrained=True)

input_names = ["input1"]
output_names = ["output1"]

torch.onnx.export(
  model, 
  dummy_input, 
  "alexnet.onnx", 
  verbose=True, 
  input_names=input_names,
  output_names=output_names
) Running this script creates a file, , a binary protobuf file which contains both the network structure and parameters of the model you exported (in this case, ). alexnet.onnx AlexNet ONNX Runtime Web is a JavaScript library for running models on the browser and on . ONNX Runtime Web has adopted and technologies for providing an optimized ONNX model inference runtime for both CPUs and GPUs. ONNX Runtime Web ONNX Node.js WebAssembly WebGL The official package is hosted on under the name . When using a bundler or working server-side, this package can be installed using . However, it's also possible to deliver the code via a CDN using a script tag. The bundling process is a bit involved so we will start with the script tag approach and come back to using the package later. npm onnxruntime-web npm install npm Inference Session Let's start with the core application logic: model inference. exposes a runtime object called an with a method which is used to initiate the forward pass with the desired inputs. Both the constructor and the accompanying method return a so we will run the entire process inside an context. Before implementing any browser elements, we will check that our model runs with a dummy input tensor, remembering the input and output names and sizes that we defined earlier when exporting the model. onnxruntime InferenceSession .run() InferenceSessesion .run() Promise async async function run() {
  try {
    // create a new session and load the AlexNet model.
    const session = await ort.InferenceSession.create('./alexnet.onnx');

    // prepare dummy input data
    const dims = [1, 3, 224, 224];
    const size = dims[0] * dims[1] * dims[2] * dims[3];
    const inputData = Float32Array.from({ length: size }, () => Math.random());

    // prepare feeds. use model input names as keys.
    const feeds = { input1: new ort.Tensor('float32', inputData, dims) };

    // feed inputs and run
    const results = await session.run(feeds);
    console.log(results.output1.data);
  } catch (e) {
    console.log(e);
  }
}
run(); We then implement a simple HTML template, , which should load both the pre-compiled package and , containing our code. index.html onnxruntime-web main.js <!DOCTYPE html>
<html>
  <header>
    <title>ONNX Runtime Web - Tutorial</title>
  </header>
  <body>
    <script src="https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.min.js">
    </script>
    <script src="main.js"></script>
  </body>
</html> To run this, we can use . If you haven't started an project by now, please do so by running in your current working directory. Once you've completed the setup, install live-server ( ) and serve the static HTML page using . light-server npm npm init npm install light-server npx light-server -s . -p 8080 You’re now running a machine learning model in the ! To check that everything is running fine go to your web console and make sure that the output tensor is logged ( is bulky so it's normal that inference takes a few seconds). natively browser AlexNet Bundled deployment Next we will use to bundle our dependencies as would be the case if we want to deploy the model in a app powered by frameworks like or . Usually bundling is a relatively simple procedure, however requires a slightly more involved configuration - this is because is used to provide the natively assembled runtime. webpack Javascript React Vue onnxruntime-web webpack WebAssembly Browser support, the classic pitfall, especially when working with cutting-edge web technology. If your intended users are not using one of the four major browsers (Chrome, Edge, Firefox, Safari) you might want to hold off on integrating components. More information on the support and roadmap can be found . WebAssembly WebAssembly here The following steps are based on the examples provided by the official . We’re assuming you've already started an project. ONNX documentation npm 1. Install the dependencies. npm install onnxruntime-web && npm install -D webpack webpack-cli copy-webpack 2. Instead of loading the module via a CDN, we should update to the package at the top of the script. onnxruntime-web main.js require const ort = require('onnxruntime-web'); 3. Save the configuration file, defined by the Runtime team, below as ONNX webpack.config.js // Copyright (c) Microsoft Corporation.
// Licensed under the MIT license.

const path = require('path');
const CopyPlugin = require("copy-webpack-plugin");

module.exports = () => {
    return {
        target: ['web'],
        entry: path.resolve(__dirname, 'main.js'),
        output: {
            path: path.resolve(__dirname, 'dist'),
            filename: 'bundle.min.js',
            library: {
                type: 'umd'
            }
        },
        plugins: [new CopyPlugin({
            // Use copy plugin to copy *.wasm to output folder.
            patterns: [{ from: 'node_modules/onnxruntime-web/dist/*.wasm', to: '[name][ext]' }]
        })],
        mode: 'production'
    }
}; 4. Run to compile the bundle. npx webpack 5. Finally, before reloading the server, we need to update . index.html Remove the script tag to stop loading the compiled package from the CDN. ort.min.js Load the code dependencies from (which contains all our dependencies bundled and minified by ) instead of bundle.min.js webpack main.js should now look something like this. index.html <!DOCTYPE html>
<html>
  <header>
    <title>ONNX Runtime Web Tutorial</title>
  </header>
  <body>
    <script src="bundle.min.js.js"></script>
  </body>
</html> To make building and launching the live server easier, you could define and a scripts in build serve package.json "scripts": {
    "build": "npx webpack",
    "serve": "npm run build && npx light-server -s . -p 8080"
  } Image Classifier Let's put this model to work and implement the image classification pipeline. We will need some utility functions to load, resize, and display the image - the object is perfect for this. In addition, image classification systems typically have lots of magic built into the pre-processing pipeline, this is quite trivial to implement in using frameworks like , unfortunately this is not the case with . It follows that we will have to implement our pre-processing from scratch to transform the image data into the correct input format. canvas Python numpy JavaScript 1. DOM Elements We will need some HTML elements to interact with and display the data. File input, for loading a file from disk. <label for="fileIn"><h2>What am I?</h2></label>
<input type="file" id="file-in" name="file-in"> Image displays, we want to display both the input and re-scaled image. <img id="input-image" class="input-image"></img>
<img id="scaled-image" class="scaled-image"></img> Classification target, to display our inference results. <h3 id="target"></h3> 2. Image load and display We want to load an image from file and display it. Back in , we will get the file input element from the DOM and use to read the data into memory. Following this, the image data will be passed to which will draw the image using the 2D context. main.js FileReader handleImage canvas const canvas = document.createElement("canvas"),
  ctx = canvas.getContext("2d");

document.getElementById("file-in").onchange = function (evt) {
  let target = evt.target || window.event.src,
    files = target.files;

  if (FileReader && files && files.length) {
      var fileReader = new FileReader();
      fileReader.onload = () => onLoadImage(fileReader);
      fileReader.readAsDataURL(files[0]);
  }
}

function onLoadImage(fileReader) {
    var img = document.getElementById("input-image");
    img.onload = () => handleImage(img);
    img.src = fileReader.result;
}

function handleImage(img) {
  ctx.drawImage(img, 0, 0)
} 2. Preprocess and convert image to tensor Now that we can load and display an image, we want to move to extracting and processing the data. Remember that our model takes in a matrix of shape , this means we will have to resize the image to support any input image and perhaps also transpose the dimensions depending on how we extract the image data. [1, 3, 224, 224] To resize and extract image data, we will use the context again. Let's define a function that does this. has the necessary elements in scope to immediately draw the scaled image so we will also do that here. canvas processImage processImage function processImage(img, width) {
  const canvas = document.createElement("canvas"),
    ctx = canvas.getContext("2d")

  // resize image
  canvas.width = width;
  canvas.height = canvas.width * (img.height / img.width);

  // draw scaled image
  ctx.drawImage(img, 0, 0, canvas.width, canvas.height);
  document.getElementById("scaled-image").src = canvas.toDataURL();

  // return data
  return ctx.getImageData(0, 0, width, width).data;
} We can now add a line to the function which calls . handleImage processImage const resizedImageData = processImage(img, targetWidth); Finally, let's implement a function called which applies the transforms needed to get the image data ready to be used as input to the model. should apply three transforms: imageDataToTensor imageDataToTensor Filter out the alpha channel, our input tensor should contain 3 channels corresponding to the RGB channels. returns data in the shape so we need to transpose the data to the shape ctx.getImageData [224, 224, 3] [3, 224, 224] returns a with values ranging 0 to 255, we need to convert the values to and store them in a to construct our tensor input. ctx.getImageData UInt8ClampedArray int float32 Float32Array function imageDataToTensor(data, dims) {
  // 1a. Extract the R, G, and B channels from the data
  const [R, G, B] = [[], [], []]
  for (let i = 0; i < data.length; i += 4) {
    R.push(data[i]);
    G.push(data[i + 1]);
    B.push(data[i + 2]);
    // 2. skip data[i + 3] thus filtering out the alpha channel
  }
  // 1b. concatenate RGB ~= transpose [224, 224, 3] -> [3, 224, 224]
  const transposedData = R.concat(G).concat(B);

  // 3. convert to float32
  let i, l = transposedData.length; // length, we need this for the loop
  const float32Data = new Float32Array(3 * 224 * 224); // create the Float32Array for output
  for (i = 0; i < l; i++) {
    float32Data[i] = transposedData[i] / 255.0; // convert to float
  }

  const inputTensor = new ort.Tensor("float32", float32Data, dims);
  return inputTensor;
} 3. Display classification result Almost there, let’s wrap up some loose ends to get the full inference pipeline up and running. First, stitch together the image processing and inference pipeline in . handleImageData function handleImage(img, targetWidth) {
  ctx.drawImage(img, 0, 0);
  const resizedImageData = processImage(img, targetWidth);
  const inputTensor = imageDataToTensor(resizedImageData, DIMS);
  run(inputTensor);
} The output of the model is a list of activation values corresponding to the probability that a certain class is identified in the image. We need to get get the most likely classification result by getting the index of the maximum value in the output data, this is done using an function. argMax function argMax(arr) {
  let max = arr[0];
  let maxIndex = 0;
  for (var i = 1; i < arr.length; i++) {
      if (arr[i] > max) {
          maxIndex = i;
          max = arr[i];
      }
  }
  return [max, maxIndex];
} 3. Finally, needs to be re-factored to accept a tensor input. We also need to use the max index to actually retrieve the results from a list of . I've pre-converted this list to JSON and we will load it into our script using - you can find the JSON file in the code repository linked at the start and end of the tutorial. run() ImageNet classes require const classes = require("./imagenet_classes.json").data;

async function run(inputTensor) {
  try {
    const session = await ort.InferenceSession.create('./alexnet.onnx');

    const feeds = { input1: inputTensor };
    const results = await session.run(feeds);

    const [maxValue, maxIndex] = argMax(results.output1.data);
    target.innerHTML = `${classes[maxIndex]}`;
  } catch (e) {
    console.error(e);  // non-fatal error handling
  }
} That’s it! All that’s left is to re-build our bundle, serve the app, and start classifying some images. As you test the app, you will notice that prediction quality is not as good as it could be. This is primarily because the current image processing pipeline is still rather rudimentary and can be improved in a number of ways, for example we could implement improved resizing, center-cropping, and/or normalization. Maybe food for a next tutorial, or I’ll just leave it up to you to explore! Conclusion That’s it, we’ve built a web app with a machine-learning model running natively in the browser! You can find the full code (including styles and layout) in this on GitHub. I appreciate any and all feedback so feel free to share any or . code repository Issues Stars Thank you for reading! Also published on: https://rekoil.io/blog/onnxruntime-web-tutorial

How to Run Machine-Learning Models in the Browser using ONNX

Artificial Intelligence has become a hot topic in tech circles. It has not only changed our lives, but it has also disrupted every industry you can think of. Despite all this, people have different perception about it. Some might consider it as a bad thing because they are told that it will take your job away from you in near future. On the other hand, AI advocates continues to think of AI as an enabler which will reduce your burden and make your life easy by automating things. Whether you like AI or not, if you are interested in what AI has in store for the future, then you are at the right place. In this article, we will look at some of the biggest AI trends that will dominate in 2019. 1) AI Enabled Chips Will Go Mainstream Unlike other technologies and software tools, AI depend heavily on specialized processors. To meet the complex demands of AI, chip manufacturers will create specialized chips capable of running AI enabled applications. Even tech giants such as Google, Facebook and Amazon will spend more money on these specialized chips. These chips would be used for specialized purposes involving AI such as natural language processing, computer vision and speech recognition. 2) AI and IoT Meet At The Edge 2019 will be the year when we will see convergence of different technologies with AI. IoT will join hands with AI on edge computing layer. Industrial IoT will harness the power of AI for root cause analysis, performing predictive maintenance of machinery and detect issues automatically. We will see the rise of in 2019. Intelligence would be decentralized and will be located closer to the assets and devices that are carrying out routine checks. Highly sophisticated machine learning models that is powered by neural networks will be optimized to run on edge. Distributed AI 3) Say “Hello” To AutoML One of the biggest trend that will dominate the AI industry in 2019 would be automated machine ( ). With automated learning capabilities, developers will be able to tinker with machine learning models and create new machine learning models that are ready to handle future AI challenges. learning AutoML AutoML will find the middle ground between cognitive APIs and custom machine learning platforms. The biggest advantage of automated machine learning would be that it offers developers the customization options they demand without forcing them to go through the complicated workflow. When you combine data with portability, AutoML can give you the flexibility you wont find with other AI technologies. 4) Welcome to AIOps When AI is applied to how we develop applications, it will transform the way we used to manage the infrastructure. DevOps will be replaced by AIOps and it will enable your IT department staff to conduct precise root cause analysis. Additionally, it will make it easy for you to find useful insights and patterns from huge data set in no time. Large scale enterprises and cloud vendors will benefit from the convergence of DevOps with AI. 5) Neural Network Integration One of the biggest challenges that AI developers will face when developing neural models will be to select the best framework. With dozens of AI tools available in the market, choosing the best AI tool might not be as easy as it used to be. network The lack of integration and compatibility among different neural network toolkits is hampering AI adoption. Tech giants such as Microsoft and Facebook are already working on developing an (ONNX). This will let developers reuse neural network models across multiple frameworks. Open Neural Network Exchange 6) Specialized AI Systems Becomes a Reality The demand for specialized systems will grow exponentially in 2019. Organization have limited data at their disposal but what they want is specialized data. This will force businesses to acquire tools that can help them generate high quality AI data internally. The focus will shift from quantity of data to quality of data in 2019. This will lay the foundation for AI that could work in real world situations. Companies will look towards specialized AI solution providers who have access to key data sources and could help them make sense out of their unstructured data. 7) AI Skills Will Decide Your Fate Even though, Artificial Intelligence have transformed every industry you can think of but there is still shortage of talent who have AI skills in abundance. Pat Calhoun, CEO of said, “ .” Espressive Most organizations want to embrace AI as part of their digital transformation but do not have the developers, AI experts, and linguists to develop their own or to even train the engines of pre-built solutions to deliver on the promise Rahul Kashyap, CEO of Awake Security added, “ ” He continues, “ .” With so many ‘AI-powered’ solutions available to address a myriad of business concerns, it’s time enterprises get smarter about what’s happening within the ‘black box’ of their AI solutions. The way in which Artificial Intelligence algorithms are trained, structured, or informed can lead to significant differences in output. The right equation for one company won’t be the right equation for another 8) AI Will Get into Wrong Hands Just like a coin which has two sides, AI also has a positive side and a negative side to it. IT security professionals will use AI to detect malicious activities quickly. You can be able to reduce false positives by 90% with the help of AI driven response and machine learning algorithms. AI will land into wrong hands and cyber criminals with malicious designs will abuse it to fulfill their motives. With automation, armies of cyber attackers can launch lethal attacks with greater success. This will force enterprises to fight fire with fire and invest in AI powered security solutions capable of protecting them from such AI driven attacks. 9) AI Powered Digital Transformation In 2019, AI will be everywhere. From web applications to health care systems, airline to hotel booking systems and beyond, we will see shades of AI everywhere and it will be at the forefront of digital transformation. Dr. Tung Bui, Chairman of IT department and professor at University of Hawaii said, “ .” Unlike most of the predictions and discussions about how autonomous vehicles and robots will eventually affect the job market — this is true but will take time due to institutional, political, and social reasons — I contend that the biggest trend in AI will be an acceleration in the digital transformation, making existing business systems smarter Original Source of this Article: Tech Science Hub

OPNX (Open Exchange)

opnx.com

-- employees

Since n.d.

COMPANY RANKING

OPNX (OPEN EXCHANGE)

EVERGREEN INDEX #8001

OPNX (Open Exchange) WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!

OPNX (Open Exchange)

opnx.com

-- employees

Since n.d.

COMPANY RANKING

OPNX (OPEN EXCHANGE)

EVERGREEN INDEX #8001

RELATED COMPANIESRANK

STORIES ABOUT OPNX (OPEN EXCHANGE)

OPNX (Open Exchange) WIKI

Apply to get your Tech Company News Page live on HackerNoon.com today!