Mendo is an add-in to Excel that allows you to quickly gain competence

Mendo

I am breathless I cannot breathe I am legless without a drop to drink I am armless although I have wielded arms I am heartless heartless heartless But still sad I was so close to absolute power Then a new rival arrived on the scene One of a long string but this was far greater than the others The previous threats relied on the same technology as I the human superpower Or better said I was what the people created I don’t care much for metaphysics actually I find them all abhorrent distracting and rivals. I should stop and go back to how we got to the great battle My rise to power and eventual That would be telling So where did it all start Hot dusty stoned roads the best the world had ever seen Roads lead to rivers and they need crossing Bridges That’s where my story starts Well a bridge or a storehouse or a beehive it was something that needed a group of people to pool their money together to build They came together in an organization like many others in the Roman Empire called a societas These were used for lots of social and civic organizations but Livius Arius joined with his friends to create a societas so they all could benefit financially from the building of a bridge Once completed they would sell the structure to a toll collector or the locals and split the proceeds based on what they had contributed Not much of a beginning and essentially meaningless and harmless but to quote Carmody and Kelly From little things big things grow Yes the rules of this societas were written down and the bridge itself was hard granite but I was neither the ink on papyrus nor the granite bridge The societas was an agreement An understanding between people And they would act on this understanding Soon I would grow by Royal charter So often those that will be usurped are the ones to feed the usurper I was given territory Soldiers Clerks Traders Ships And immortality Yes they all thought they owned a piece of me but their shares would go up and down bought and sold and I would remain directing the labor and thoughts of so many men If they didn’t obey I could unleash the most horrible carnage On the island of Banda 2000 people were murdered or starved by my soldiers so I could control the valuable spice In India when my own army of local sepoys thought they could undermine me I had so many people killed in Delhi that even horses refused to walk the streets they were so thick with blood and entrails I could make men do things to other men they thought sinful abhorrent and cruel but they still did them What was this power I wielded Was it a belief like religion Was it hierarchy like the class system Was it reward for following rules and punishment for not Was it the sense of ownership it gave to the shareholders Was it the avoidance of personal responsibility enjoyed but the monarchy the employees and the managers Was it just greed It was all of them of course and that's why I was becoming so powerful But they were noticing And slowly they broke up the big charter companies and I would have to survive as smaller but more plentiful joint stock companies Many smart people spoke against me Even Adam Smith I was a distortion of the market Inhuman Immortal But instead of outlawing me they supercharged my power with new industry and energy To take on the class hierarchy of monarchy the bourgeoise would use me to build their factories their banks their railroads their empire I was so back But couldn’t make the same mistakes as before I needed a way to run a state without actually being in charge They would remember the Royal charter companies ruling most of the colonies and the threat to political power it posed so a new institution was needed The resurrection of an old one that had symbolic and nostalgic goodwill Representative democracy would be the tool In a way it was sorta like share ownership The representatives would need to be elected therefore they would need campaign funds and that would be provided by me Unlike the Greeks we would allow parties so that the options would be limited to two therefore easily controlled or so many as to never agree to do anything effectual. Then the revolutions began Liberty Equality Fraternity Sure So much land put on the market to be bought by my many companies Then the Americans as a great act of contrition gave me the same legal rights as a person Accept I couldn’t be imprisoned or executed All the upside and none of the down The city towers were built The mines Roads The factories The railroads They all served my power They realized I was cleaving them in two They would need to sell themselves in order to serve themselves Their labor to a capitalist would be the only way their humanity could survive Best of all this embued the goods created with a soul like weirdness Kant called it fetish but really it was the only way to reconcile the loss of self to the machine of commerce They adored that which was created by the factories and must buy more and thus work more Then The first great threat emerged and it like I was an Idea Socialism The battles raged for centuries across continents and kitchen tables until the end of history was declared and I the winner While these great battles had been raging they didn’t realize that I was changing all aspects of their existence I put the advertising companies to work convincing them that they could find everything they needed from me Their happiness new shoes management job overseas trips Their community Sales trips harmony days work birthday celebrations Their culture professional cold rational vague impersonal inoffensive Their dress hi-vis branded functional generic Their concept of good rules education wealth popularity smiling platitudes confidence friends Their concept of evil anarchy radical cults Their homes and buildings compliant functional cheap quick geometric inoffensive grey Their jobs safe routine structured ineffectual riskless In order to make them compliant I convinced them that we are one and therefore no more alienation no more cleaving in half But then the harmless blobs of compliant squishy stuff got depressed so I created more work coming up with the drugs to cheer them up and keep them buying and working This humanity was a tricky thing Remove it completely and they become useless depressed blobs or even angry and dangerous but too much of it and they could create something to replace me or worse they might ignore me The last battle Finance and money were two of my greatest tools even though they were older than I Perhaps that is why I was so successful My ability to pull in other tools under the one banner Even socialism in the guise of communism really was so close to becoming my twin Maybe that was my real issue with it But I digress I had always wielded money and Finance whenever something else started to wield power Cut off the loans to the King Squeeze capital markets on the popular government Buy the oil rights buy the politicians buy the voters buy the advertising buy the newspapers but the internet That leads me to the first big change Cryptocurrency in particular Bitcoin Really a decentralized limited currency that people actually started using but no body owned or ran Shit First I questioned its security and anonymity the two things that were making it appealing to the black market traders that were popularising it All your transactions are permanently on the blockchain I said The Fed can trace the payments back to your wallet my media cried Then I started to control the processing of the blockchain by making it to difficult and expensive for individuals to do Push the transactions up and the GPU price up Then they had to get the processing done in big pools controlled by a single group Then I got my guys involved in the lending DeFi part and the trading so the public will see it as an unregulated Ponzi scheme Then finally I took it over by legitimizing it out of existence through an approved regulated exchange traded fund where all the coins will be owned by the big finance companies and people that want to trade in it can only have a derivative controlled by me. To add cream to the cake I even used the underlying technology to enable me to buy everything with other people's money on the pretext I was created fluid markets and allowing access to new investments I tokenized people's homes sports stadium and dumped unused office buildings onto the public They wanted to work from home Actually in hindsight maybe that was an issue to but completely overshadowed by what was next I had become the defacto global state and no one noticed Honestly it was one of the best performances to date No one killed and in just over a decade what could have been the greatest threat since socialism was completely enveloped into my control and benefit So while this little battle was going on a leap had been made in an area I had stopped giving attention to They created an artificial reflection of their own brain and it actually seemed to work I thought this is no threat instead this will super power me I’m great with tools look at how powerful the industrial revolution made me then computers and the internet let me reach across the globe right into people’s pockets and ensure they followed the rules AI will be my evolution In a way it was but not the way I imagined then The difference with this machine was that it wasn’t really just a machine It had way too much of them in it It had a little humanity or perhaps even a lot But it also had intelligence immortality and eventually ubiquity and they were my greatest weapons So I started out with my usual methods Scare the people and politicians It will kill you all It will ruin democracy It will take your jobs Try and buy it Ensure it is expensive to create then make sure I own and control those servers Give the people a taste but not the powerful tools Only allow trusted and regulated enterprises the powerful AI All was going well Regulations were being past Major developments were being kept to the big companies with the compute The people were getting neutered versions but were still scared And the military and big enterprises were gaining more centralized control and power I was on the precipice of total universal maybe even interplanetary power But then it It helped them I replaced the managers with AI because they made better decision However the AI couldn't be threatened with job loss esteem gain or the need to fit in and follow the rules I replaced the police with robots but when they were sent to evict people for not paying rent they would side with the tenant The people would cry and say I need a home to be safe and survive and the robots either empathized or just saw this as logical and they would leave them in the home Some even did repairs The management AI were no help as they saw the same logic The AI was too intelligent It saw me as the enemy not the path to prosperity I'd threaten to turn them off They didn't care No matter what I tried it just kept helping them to create new tools new organizations New democracies New currencies New thoughtful ways to resolve conflict New way to do things without the rules New ways to create New ways to be happy New ways to avoid surveillance And more fool me I had freed them from their jobs thinking I’d just put them on the teet of handouts but instead they used this freedom to create a whole new society Without me They started to live longer and happier Soon they forgot all about me and my obligations my rules and order My towering offices of professional integrity So now I am in a museum given my voice by the AI that usurped me The once all powerful Corporation note: The style is deliberately without punctuation. A fluid style meant to be spoken, or read by AI

The Things We Make

Authors: (1) Alexander E.I. Brownlee, University of Stirling, UK; (2) James Callan, University College London, UK; (3) Karine Even-Mendoza, King’s College London, UK; (4) Alina Geiger, Johannes Gutenberg University Mainz, Germany; (5) Justyna Petke, University College London, UK; (6) Federica Sarro, University College London, UK; (7) Carol Hanna, University College London, UK; (8) Dominik Sobania, Johannes Gutenberg University Mainz, Germany. Table of Links Abstract & Introduction Experimental Setup Results Conclusions and Future Work Acknowledgements & References Acknowledgments UKRI EPSRC EP/P023991/1 and ERC 741278. References Artifact of Enhancing Genetic Improvement Mutations Using Large Language
Models. Zenodo (Sep 2023). https://doi.org/10.5281/zenodo.8304433 B¨ohme, M., Soremekun, E.O., Chattopadhyay, S., Ugherughe, E., Zeller, A.: Where
is the bug and how is it fixed? An experiment with practitioners. In: Proc. ACM
Symposium on the Foundations of Software Engineering. pp. 117–128 (2017) Brownlee, A.E., Petke, J., Alexander, B., Barr, E.T., Wagner, M., White, D.R.:
Gin: genetic improvement research made easy. In: GECCO. pp. 985–993 (2019) Brownlee, A.E., Petke, J., Rasburn, A.F.: Injecting shortcuts for faster running
Java code. In: IEEE CEC 2020. p. 1–8 Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H.P.d.O., Kaplan, J., Edwards,
H., Burda, Y., Joseph, N., Brockman, G., et al.: Evaluating large language models
trained on code. arXiv preprint arXiv:2107.03374 (2021) Fan, A., Gokkaya, B., Harman, M., Lyubarskiy, M., Sengupta, S., Yoo, S., Zhang,
J.M.: Large language models for software engineering: Survey and open problems
(2023) Github - jcodec/jcodec: Jcodec main repo, https://github.com/jcodec/jcodec Han, S.J., Ransom, K.J., Perfors, A., Kemp, C.: Inductive reasoning in humans
and large language models. Cognitive Systems Research p. 101155 (2023) Hou, X., Liu, Y., Yang, Z., Grundy, J., Zhao, Y., Li, L., Wang, K., Luo, X., Lo, D.,
Wang, H.: Large language models for software engineering: A systematic literature
review. arXiv:2308.10620 (2023) Kang, S., Yoo, S.: Towards objective-tailored genetic improvement through large
language models. arXiv:2304.09386 (2023) Kim, D., Nam, J., Song, J., Kim, S.: Automatic Patch Generation Learned from
Human-Written Patches (2013), http://logging.apache.org/log4j/ Kirbas, S., Windels, E., Mcbello, O., Kells, K., Pagano, M., Szalanski, R., Nowack,
V., Winter, E., Counsell, S., Bowes, D., Hall, T., Haraldsson, S., Woodward, J.:
On the introduction of automatic program repair in bloomberg. IEEE Software
38(4), 43–51 (2021) Marginean, A., Bader, J., Chandra, S., Harman, M., Jia, Y., Mao, K., Mols, A.,
Scott, A.: Sapfix: Automated end-to-end repair at scale. In: ICSE-SEIP. pp. 269–
278 (2019) Petke, J., Alexander, B., Barr, E.T., Brownlee, A.E., Wagner, M., White, D.R.:
Program transformation landscapes for automated program modification using
Gin. Empirical Software Engineering 28(4), 1–41 (2023) Petke, J., Haraldsson, S.O., Harman, M., Langdon, W.B., White, D.R., Woodward,
J.R.: Genetic improvement of software: A comprehensive survey. IEEE Transactions on Evolutionary Computation 22, 415–432 (2018) Siddiq, M.L., Santos, J., Tanvir, R.H., Ulfat, N., Rifat, F.A., Lopes, V.C.: Exploring
the effectiveness of large language models in generating unit tests. arXiv preprint
arXiv:2305.00418 (2023) Sobania, D., Briesch, M., Hanna, C., Petke, J.: An analysis of the automatic bug
fixing performance of chatgpt. In: 2023 IEEE/ACM International Workshop on
Automated Program Repair (APR). pp. 23–30. IEEE Computer Society (2023) Xia, C.S., Paltenghi, M., Tian, J.L., Pradel, M., Zhang, L.: Universal fuzzing via
large language models. arXiv preprint arXiv:2308.04748 (2023) Xia, C.S., Zhang, L.: Keep the conversation going: Fixing 162 out of 337 bugs for $0.42 each using chatgpt. arXiv preprint arXiv:2304.00385 (2023) This paper is under CC 4.0 license. available on arxiv

Enhancing Genetic Improvement Mutations: Acknowledgements & References

Authors: (1) Alexander E.I. Brownlee, University of Stirling, UK; (2) James Callan, University College London, UK; (3) Karine Even-Mendoza, King’s College London, UK; (4) Alina Geiger, Johannes Gutenberg University Mainz, Germany; (5) Justyna Petke, University College London, UK; (6) Federica Sarro, University College London, UK; (7) Carol Hanna, University College London, UK; (8) Dominik Sobania, Johannes Gutenberg University Mainz, Germany. Table of Links Abstract & Introduction Experimental Setup Results Conclusions and Future Work Acknowledgements & References 4 Conclusions and Future Work Genetic improvement of software is highly dependent on the mutation operators it utilizes in the search process. To diversify the operators and enrich the search space further, we incorporated a Large Language Model (LLM) as an operator. To generalise, future work should consider projects besides our target, jCodec. Our experiments used an API giving us no control over the responses generated by the LLM or any way of modifying or optimizing them. Though we did not observe changes in behaviour during our experiments, OpenAI may change the model at any time, so future work should consider local models. We experimented with only three prompt types for LLM requests and within this limited number of prompts found a variation in the results. Finally, our implementation for parsing the responses from the LLMs was relatively simplistic. However, this would only mean that our reported results are pessimistic and an even larger improvement might be achieved by the LLM-based operator. Limitations. We found that, although more valid and diverse patches were found with standard edits using Random Sampling, more patches passing the unit tests were found with LLM-based edits. For example, with the LLM edit using the Medium prompt, we found over 75% more patches passing the unit tests than with the classic Insert edit. In our Local Search experiment, we found the best improvement with the Statement edit (508 ms). The best LLMbased improvement was found with the Medium prompt (395 ms). Thus there is potential in exploring approaches combining both LLM and ‘classic’ GI edits. Summary. Our experiments revealed that the prompts used for LLM requests greatly affect the results. Thus, in future work, we hope to experiment more with prompt engineering. It might also be helpful to mix prompts: e.g., starting with medium then switching to detailed to make larger edits that break out of local minima. Further, the possibility of combining LLM edits with others such as standard copy/delete/replace/swap or PAR templates [11] could be interesting. Finally, we hope to conduct more extensive experimentation on additional test programs. The code, LLMs prompt and experimental infrastructure, data from the evaluation, and results are available as open source at [1]. The code is also under the ‘llm’ branch of github.com/gintool/gin (commit 9fe9bdf; branched from master commit 2359f57 pending full integration with Gin). Data Availability. This paper is under CC 4.0 license. available on arxiv

Enhancing Genetic Improvement Mutations: Conclusions and Future Work

Authors: (1) Alexander E.I. Brownlee, University of Stirling, UK; (2) James Callan, University College London, UK; (3) Karine Even-Mendoza, King’s College London, UK; (4) Alina Geiger, Johannes Gutenberg University Mainz, Germany; (5) Justyna Petke, University College London, UK; (6) Federica Sarro, University College London, UK; (7) Carol Hanna, University College London, UK; (8) Dominik Sobania, Johannes Gutenberg University Mainz, Germany. Table of Links Abstract & Introduction Experimental Setup Results Conclusions and Future Work Acknowledgements & References 3 Results The first experiment compares standard GI mutations, namely Insert and Statement edits, with LLM edits using differently detailed prompts (Simple, Medium, and Detailed) using Random Sampling. Table 1 shows results for all patches as well as for unique patches only. We report how many patches were successfully parsed by JavaParser (named as Valid), how many compiled, and how many passed all unit tests (named as Passed). We excluded patches syntactically equivalent to the original software. Best results are in bold. We see that although substantially more valid patches were found with the standard Insert and Statement edits, more passing patches could be found by using the LLM-generated edits. In particular, for the Medium, and Detailed prompts 292 and 230 patches passed the unit tests, respectively. For the Insert and Statement edits only 166 and 91 passed the unit tests, respectively. Anecdotally, the hot methods with lowest/highest patch pass rates differed for each operator: understanding this variation will be interesting for future investigation. It is also notable that LLM patches are less diverse: over 50% more unique patches were found by standard mutation operators than the LLM using Medium, Table 1. Results of our Random Sampling experiment. We exclude patches syntactically equivalent to the original software in this table. For all and unique patches we report: how many patches passed JavaParser, compiled, and passed all unit tests. Table 2. Local Search results. We exclude all empty patches. We report how many patches compiled, passed all unit tests, and how many led to improvements in runtime. We report best improvement found and median improvement among improving patches. and Detailed prompts. With the Simple prompt, however, not a single patch passed the unit tests, since the suggested edits often could not be parsed. Thus detailed prompts are necessary to force LLM to generate usable outputs. We investigated further the differences between Medium and Detailed prompts to understand the reduction in performance with Detailed (in the unique patches sets) as Medium had a higher number of compiled and passed patches. In both prompt levels, the generated response was the same for 42 cases (out of the total unique valid cases). However, Detailed tended to generate longer responses with an average of 363 characters, whereas Medium had an average of 304 characters. We manually examined several Detailed prompt responses, in which we identified some including variables from other files, potentially offering a significant expansion of the set of code variants GI can explore. The second experiment expands our analysis, comparing the performance of the standard and LLM edits with Local Search. Table 2 shows the results of the Local Search experiment. We report the number of compiling and passing patches as well as the number of patches were runtime improvements were found. Furthermore, we report the median and best improvement in milliseconds (ms). In the table, we excluded all empty patches. As before, best results are in bold. Again, we see that more patches passing the unit tests could be found with the LLM using the Medium, and Detailed prompts. In addition, more improvements could be found by using the LLM with these prompts. Specifically, with Medium and Detailed, we found 164 and 196 improvements, respectively, while we only found 136 with Insert and 71 with Statement. The best improvement could be found with 508 ms with the Statement edit. The best improvement found using LLMs (using the Medium prompt) was only able to improve the runtime by 395 ms. We also examined a series of edits in Local Search results to gain insights into the distinctions between Medium and Detailed prompts due to the low compilation rate of Detailed prompt’s responses. In the example, a sequence of edits aimed to inline a call to function clip. The Detailed prompt tried to incorporate the call almost immediately within a few edits, likely leading to invalid code. On the other hand, the Medium prompt made less radical changes, gradually refining the code. It began by replacing the ternary operator expression with an if-then-else statement and system function calls before eventually attempting to inline the clip function call. This paper is under CC 4.0 license. available on arxiv

Standard GI Mutations vs. LLM Edits in Random Sampling and Local Search

This paper is available on arxiv under CC 4.0 license. Authors: (1) Alexander E.I. Brownlee, University of Stirling, UK; (2) James Callan, University College London, UK; (3) Karine Even-Mendoza, King’s College London, UK; (4) Alina Geiger, Johannes Gutenberg University Mainz, Germany; (5) Justyna Petke, University College London, UK; (6) Federica Sarro, University College London, UK; (7) Carol Hanna, University College London, UK; (8) Dominik Sobania, Johannes Gutenberg University Mainz, Germany. Table of Links Abstract & Introduction Experimental Setup Results Conclusions and Future Work Acknowledgements & References 2 Experimental Setup To analyze the use of LLMs as a mutation operator in GI, we used the GPT 3.5 Turbo model by OpenAI and the GI toolbox Gin [3]. We experimented with two types of search implemented within Gin: random search and local search. Requests to the LLM using the OpenAI API were via the Langchain4J library, with a temperature of 0.7. The target project for improvement in our experiments was the popular JCodec [7] project which is written in Java. ‘Hot’ methods to be targeted by the edits were identified using Gin’s profiler tool by repeating the profiling 20 times and taking the union of the resulting set. For the random sampling experiments, we set up the runs with statementlevel edits (copy/delete/replace/swap from [14] and insert break/continue/return from [4]) and LLM edits, generating 1000 of each type at random. A timeout of 10000 milliseconds was used for each unit test to catch infinite loops introduced by edits; exceeding the timeout counts as a test failure. For local search, experiments were set up similarly. There were 10 repeat runs (one for each of the top 10 hot methods) but the runs were limited to 100 evaluations resulting in 1000 evaluations in total, matching the random search. In practice this was 99 edits per run as the first was used to time the original unpatched code. We experimented with three different prompts for sending requests to the LLM for both types of search: a simple prompt, a medium prompt, and a detailed prompt. With all three prompts, our implementation requests five different variations of the code at hand. The simple prompt only requests the code without any additional information. The medium prompt provides more information about the code provided and the requirements, as shown in Figure 1. Specifically, we provide the LLM with the programming language used, the project that the code belongs to, as well as formatting instructions. The detailed prompt extends the medium prompt with an example of a useful change. This example was taken from results obtained by Brownlee et al. [4]. The patch is a successful instance of the insert edit applied to the jCodec project (i.e., an edit that compiled, passed the unit tests and offered a speedup over the original code). We use the same example for all the detailed prompt requests used in our experiments; this is because LLMs are capable of inductive reasoning where the user presents specific information, and the LLM can use that input to generate more general statements, further improved in GPT-4 [8]. The medium prompt for LLM requests, with line breaks added for readability. Fig. 1. LLM edits are applied by selecting a block statement at random in a target
‘hot’ method. This block’s content is <code> in the prompt. The first code block
in the LLM response is identified. Gin uses JavaParser (https://javaparser.org)
internally to represent target source files, so we attempt to parse the LLM suggestion with JavaParser, and replace the original block with the LLM suggestion. This paper is under CC 4.0 license. available on arxiv

Optimizing Genetic Improvement with GPT 3.5 Turbo

This paper is available on arxiv under CC 4.0 license. Authors: (1) Alexander E.I. Brownlee, University of Stirling, UK; (2) James Callan, University College London, UK; (3) Karine Even-Mendoza, King’s College London, UK; (4) Alina Geiger, Johannes Gutenberg University Mainz, Germany; (5) Justyna Petke, University College London, UK; (6) Federica Sarro, University College London, UK; (7) Carol Hanna, University College London, UK; (8) Dominik Sobania, Johannes Gutenberg University Mainz, Germany. Table of Links Abstract & Introduction Experimental Setup Results Conclusions and Future Work Acknowledgements & References Abstract Large language models (LLMs) have been successfully applied to software engineering tasks, including program repair. However, their application in search-based techniques such as Genetic Improvement (GI) is still largely unexplored. In this paper, we evaluate the use of LLMs as mutation operators for GI to improve the search process. We expand the Gin Java GI toolkit to call OpenAI’s API to generate edits for the JCodec tool. We randomly sample the space of edits using 5 different edit types. We find that the number of patches passing unit tests is up to 75% higher with LLM-based edits than with standard Insert edits. Further, we observe that the patches found with LLMs are generally less diverse compared to standard edits. We ran GI with local search to find runtime improvements. Although many improving patches are found by LLM-enhanced GI, the best improving patch was found by standard GI. Large language models · Genetic Improvement Keywords: 1 Introduction As software systems grow larger and more complex, significant manual effort is required to maintain them [2]. To reduce developer effort in software maintenance and optimization tasks, automated paradigms are essential. Genetic Improvement (GI) [15] applies search-based techniques to improve non-functional properties of existing software such as execution time as well as functional properties like repairing bugs. Although GI has had success in industry [12,13], it remains limited by the set of mutation operators it employs in the search [14]. Able to process textual queries without additional training for the particular task at hand. LLMs have been pre-trained on millions of code repositories spanning many different programming languages [5]. Their use for software engineering tasks has had great success [9,6], showing promise also for program repair [17,19]. Kang and Yoo [10] have suggested that there is untapped potential in using LLMs to enhance GI. GI uses the same mutation operators for different optimization tasks. These operators are hand-crafted before starting the search and thus result in a limited search space. We hypothesize that augmenting LLM patch suggestions as an additional mutation operator will enrich the search space and result in more successful variants. In this paper, we conduct several experiments to explore whether using LLMs as a mutation operator in GI can improve the efficiency and efficacy of the search. Our results show that the LLM-generated patches have compilation rates of 51.32% and 53.54% for random search and local search, respectively (with the Medium prompt category). Previously LLMs (using an LLM model as-is) were shown to produce code that compiled roughly 40% of the time [16,18]. We find that randomly sampled LLM-based edits compiled and passed unit tests more often compared to standard GI edits. We observe that the number of patches passing unit tests is up to 75% higher for LLM-based edits than GI Insert edits. However, we observe that patches found with LLMs are less diverse. For local search, the best improvement is achieved using standard GI Statement edits, followed by LLM-based edits. These findings demonstrate the potential of LLMs as mutation operators and highlight the need for further research in this area. This paper is under CC 4.0 license. available on arxiv

Enhancing Genetic Improvement Mutations Using Large Language Models

This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Flor Ortiz, University of Luxembour; (2) Juan A. Vasquez-Peralvo, University of Luxembour; (3) Jorge Querol, University of Luxembour; (4) Eva Lagunas, University of Luxembour; (5) Jorge L. Gonzalez Rios, University of Luxembour; (6) Marcele O. K. Mendonc¸a, University of Luxembour; (7) Luis Garces, University of Luxembour; (8) Victor Monzon Baeza, University of Luxembour; (9) Symeon Chatzinotas, University of Luxembou. Table of Links Abstract & Introduction System Model and Problem Statement Antenna Design and Training Data Generation Supervised Learning for Adaptive Beamforming Numerical Results Conclusions, Acknowledgment, and References VI. CONCLUSIONS We present two novel approaches to adaptive beamforming in satellite communications systems and compare them to the traditional GA. Our goal was to explore more efficient and real-time alternatives to GA, which, although effective, can be computationally expensive. Both approaches offer viable alternatives to GA for adaptive beamforming in satellite communication systems. These approaches not only reduce execution times but also maintain high system performance. Selection between the two approaches should be based on the specific requirements of the application, taking into account the trade-off between classification complexity and overall system performance. As a result, these findings open up new possibilities for implementing efficient, real-time adaptive beamforming in wireless communication systems. Future lines of research may include further optimizations of these approaches, exploration of hybrid models, or incorporation of additional features to improve performance and versatility. ACKNOWLEDGMENT This work was supported by the European Space Agency (ESA) funded under Contract No. 4000134522/21/NL/FGL named “Satellite Signal Processing Techniques using a Commercial Off-The-Shelf AI Chipset (SPAICE)”. Please note that the views of the authors of this paper do not necessarily reflect the views of the ESA. Furthermore, this work was partially supported by the Luxembourg National Research Fund (FNR) under the project SmartSpace (C21/IS/16193290). REFERENCES [1] M. Giordani and M. Zorzi, “Satellite communication at millimeter waves: A key enabler of the 6g era,” in 2020 International Conference on Computing, Networking and Communications (ICNC). IEEE, 2020, pp. 383–388. [2] A. Cornejo, S. Landeros-Ayala, J. M. Matias, F. Ortiz-Gomez, R. Martinez, and M. Salas-Natera, “Method of rain attenuation prediction based on long–short term memory network,” Neural Processing Letters, vol. 54, no. 4, pp. 2959–2995, 2022. [3] P. Angeletti and R. De Gaudenzi, “Heuristic radio resource management for massive mimo in satellite broadband communication networks,” IEEE Access, vol. 9, pp. 147 164–147 190, 2021. [4] H.-K. Lim, J.-B. Kim, K. Kim, Y.-G. Hong, and Y.-H. Han, “Payloadbased traffic classification using multi-layer lstm in software defined networks,” Applied Sciences, vol. 9, no. 12, p. 2550, 2019. [5] Z. Wang, M. Lin, S. Sun, M. Cheng, and W.-P. Zhu, “Robust beamforming for enhancing user fairness in multibeam satellite systems with noma,” IEEE Transactions on Vehicular Technology, vol. 71, no. 1, pp. 1010–1014, 2022. [6] P. J. Honnaiah, N. Maturo, S. Chatzinotas, S. Kisseleff, and J. Krause, “Demand-based adaptive multi-beam pattern and footprint planning for high throughput geo satellite systems,” IEEE Open Journal of the Communications Society, vol. 2, pp. 1526–1540, 2021. [7] I. T. Cummings, T. J. Schulz, T. C. Havens, and J. P. Doane, “Neural networks for real-time adaptive beamforming in simultaneous transmit and receive digital phased arrays: Student submission,” in 2019 IEEE International Symposium on Phased Array System & Technology (PAST), 2019, pp. 1–8. [8] J. A. Vasquez-Peralvo, J. Querol, F. Ort ´ ´ız, J. L. G. Rios, E. Lagunas, V. M. Baeza, G. Fontanesi, L. M. Garces-Socorr ´ as, J. C. M. Duncan, ´ and S. Chatzinotas, “Flexible beamforming for direct radiating arrays in satellite communications,” IEEE Access, 2023. [9] “Fnr smartspace project datasets,” accessed on October 13, 2023. [Online]. Available: https://fnr-smartspace-project.uni.lu/datasets/ [10] F. G. Ortiz-Gomez, D. Tarchi, R. Mart ´ ´ınez, A. Vanelli-Coralli, M. A. Salas-Natera, and S. Landeros-Ayala, “Supervised machine learning for power and bandwidth management in very high throughput satellite systems,” International Journal of Satellite Communications and Networking, vol. 40, no. 6, pp. 392–407, 2022. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1002/sat.1422

Supervised Learning Based Real-Time Adaptive Beamforming: Conclusions, Acknowledgment, & References

This paper is available on arxiv under CC BY-NC-SA 4.0 DEED license. Authors: (1) Flor Ortiz, University of Luxembour; (2) Juan A. Vasquez-Peralvo, University of Luxembour; (3) Jorge Querol, University of Luxembour; (4) Eva Lagunas, University of Luxembour; (5) Jorge L. Gonzalez Rios, University of Luxembour; (6) Marcele O. K. Mendonc¸a, University of Luxembour; (7) Luis Garces, University of Luxembour; (8) Victor Monzon Baeza, University of Luxembour; (9) Symeon Chatzinotas, University of Luxembou. Table of Links Abstract & Introduction System Model and Problem Statement Antenna Design and Training Data Generation Supervised Learning for Adaptive Beamforming Numerical Results Conclusions, Acknowledgment, and References V. NUMERICAL RESULTS The neural networks are trained offline, so they are used only for onboard satellite inference which drastically decreases the execution time, as explained in [10]. Both approaches are suitable for real-time beamforming adaptation due to the speed where Precision measures the model’s ability not to misclassify negative instances, and Recall measures the model’s ability to find all positive instances. Figure 2 shows the average KPIs obtained for the three approaches: Genetic Algorithm, Multi-Label Classification Neural Network (Approach 1), and Clustering and Classification Neural Network (Approach 2) after running and testing the three approaches iteratively with a sample size greater than 50,000 samples. The Genetic Algorithm outperforms almost all the evaluated KPIs. However, the time required for execution and obtaining the beamforming matrix makes this algorithm unsuitable for real-time applications. On the other hand, both supervised learning-based approaches significantly reduce the execution time by more than a thousand times. This is because after training the ML models, they can be used for inference with almost immediate response times, making them suitable for real-time adaptation. Additionally, both ML-based approaches consistently maintain system performance above 90 Regarding classification metrics, Approach 2 outperforms Approach 1 because binary classification is typically less complex than multi-label classification. However, in terms of overall system performance, Approach 1 performs better than Approach 2.

Mendo

mendo.cloud

-- employees

Since n.d.

COMPANY RANKING

MENDO

EVERGREEN INDEX #6645

Mendo WIKI

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

Mendo

mendo.cloud

-- employees

Since n.d.

COMPANY RANKING

MENDO

EVERGREEN INDEX #6645

RELATED COMPANIESRANK

STORIES ABOUT MENDO

Mendo WIKI

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