Agenda

Dr. Erika DeBenedictis

Pioneer Labs USA

Abstract: Recent machine learning successes like Alphafold II highlight the utility of large, high-fidelity datasets in life science. Align to Innovate is a nonprofit that experiments with a new paradigm for funding, collecting, and sharing such datasets. This program will create a portfolio of robust experimental cloud lab techniques and allow scientists worldwide to submit samples for analysis. The resulting standardized measurements populate a public database after an embargo period.

Bio: Erika is a former astronomer and current synthetic biologist. As an undergraduate at Caltech, Erika worked on topics in computational physics including space mission orbit design at NASA and computational protein design at D. E. Shaw Research. Erika worked with Kevin Esvelt at MIT during her PhD, where she used laboratory automation to tackle problems in synthetic biology. Erika leads an academic group at The Francis Crick Institute in London, UK, and is the the CEO of Pioneer Labs. In 2021 she founded Align to Innovate, a nonprofit improving the reproducibility, scalability, and shareability of life science research with programmable experiments. Align hosts programs at the intersection of biology and automation, including the 2022 Bioautomation Challenge, the Protein Engineering Tournament, and the Open Datasets Initiative.

Dr. Shivagami Gugan

AWS Middle East, Turkey & Africa

Abstract: A 2024 economic study on the value of open-source software (OSS) from Harvard Business School Strategy unit estimates the supply-side value (cost to recreate the software) of widely used OSS is $4.15 billion and the demand-side value (usage) is $8.8 trillion. Firms would need to spend 3.5 times more on software than they currently do if OSS did not exist.

Amazon Web Services is proud to support open- source projects, foundations, and partners, and contributes significantly in this space. AWS also innovates faster with open source. We build on top of open source for our internal tooling as well as services such as EC2, Amazon Sagemaker, and AWS Lambda. Amazon employs dedicated engineers to work on the open-source projects with sole focus is to improve an open-source projects and the ecosystem around it.

In this talk, we will explore how AWS contributes to the open-source community, building AIML and GenAI open-source software at scale, contributing our knowledge and code back into the open-source projects we build on, and investing in the long term health of open-source communities.

Bio: Dr. Shivagami Gugan is a Technologist and Aviation Specialist with extensive experience in transforming mission-critical systems to modern architectures. Having built Organizational strategies and led the Software Engineering and Infrastructure functions at the Emirates Group for more than two decades through modernizations, she derives inspiration when technology advancements are applied rightly to deliver business value. She is a pragmatic Technology Transformation Leader who seeks to apply Data Engineering, Machine Learning, AI, Reliability Engineering and Cloud Architecture for long-standing growth and sustainability. She is a DevOps Institute lead ambassador and Linux Foundation, lead author of the SRE practitioner and the Observability courses and principal speaker at international events. At Amazon Web Services, she is the Chief Technologist for Middle East, Turkey and Africa and AI/ML/GenAI Architect for Public Sector who specializes in bringing about large-scale resilience and transformation of AI/ML/GenAI Architect for Public Sector who specializes in bringing about large-scale resilience and transformation of Government and Large Enterprises. Shivagami was awarded the "Woman of Excellence" by the Government of India for her outstanding contribution towards Technology Innovation and AI towards building the Nation.

Dr. Natalia Vassilieva

Cerebras Systems USA

Abstract: This talk will explore the various degrees of openness in AI — from access to models through inference APIs to fully open projects that share model weights, training data, details of the training recipe, source code and intermediate checkpoints. The majority of the open source models today provide only select artifacts, mostly just final model weights, while withholding data and training details. This lack of transparency limits reproducibility and complicates the development of derivative products from the open source models. We will present cases of adapting popular open source models to different languages and domains, examine the challenges posed by partial openness, and discuss approaches for overcoming these limitations. Additionally, the talk will highlight the role of efficient training techniques and scaling laws as essential tools for advancing model development.

Bio: Natalia is VP and Field CTO at Cerebras Systems. She has decades of R&D experience in NLP, CV, ML and IR. Prior to Cerebras, Natalia was a Sr. Research Manager at Hewlett Packard Labs, where she led the Software and AI group in 2015-2019 and served as the head of HP Labs Russia in 2011-2015. She led research teams developing algorithms and applications for text, image and time series analysis and modelling. In 2012-2015 Natalia also was a part-time Associate Professor at St. Petersburg State University and a part-time lecturer at Computer Science Center, St. Petersburg, Russia. She holds a PhD in CS from St. Petersburg State University.

Dr. Michal Valko

Meta Paris France

Abstract: Reinforcement learning from human feedback (RLHF) is a go-to solution for aligning large language models (LLMs) with human preferences; it passes through learning a reward model that subsequently optimizes the LLM's policy. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In the first part we turn to an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF) and give a new algorithmic solution, Nash-MD, founded on the principles of mirror descent. NLHF is compelling for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences. In the second part of the talk we delve into a deeper theoretical understanding of fine-tuning approaches as RLHF with PPO and offline fine-tuning with DPO (direct preference optimization) based on the Bradley-Terry model and come up with a new class of LLM alignment algorithms with better both practical and theoretical properties. We finish with the newest work showing links between and building on top of them.

Bio: Michal is a principal Llama engineer at Meta Paris, tenured researcher at Inria, and the lecturer at the MVA master of ENS Paris-Saclay. Michal is primarily interested in designing algorithms that would require as little human supervision as possible. That is why he is working on methods and settings that are able to deal with minimal feedback, such as deep reinforcement learning, bandit algorithms, self-supervised learning, or self play. Michal has recently worked on representation learning, word models and deep (reinforcement) learning algorithms that have some theoretical underpinning. In the past he has also worked on sequential algorithms with structured decisions where exploiting the structure leads to provably faster learning. Michal is now working on large large models (LMMs), in particular providing algorithmic solutions for their scalable fine-tuning and alignment. He received his Ph.D. in 2011 from the University of Pittsburgh under the supervision of Miloš Hauskrecht and was a postdoc of Rémi Munos before getting a tenure at Inria in 2012 and starting Google DeepMind Paris in 2018.

Dr. June Paik

FuriosaAI USA

Abstract: With the rapid rise of ChatGPT and other generative AI models, data centers are facing an explosion in demand for deep learning inference. In addition to energy efficiency, which is crucial for lowering total cost of ownership (TCO), high performance is essential for serving these large-scale models in production environments. In this presentation, we will introduce RNGD, Furiosa's 2nd gen inference AI chip, which is purpose-built to meet these challenges head-on, combining efficiency, flexibility, and cutting-edge performance to power the future of AI.

Bio: June Paik is the Founder and CEO of FuriosaAI. Before founding FuriosaAI in 2017, he worked as both a hardware and software engineer at AMD and Samsung. June holds a master’s degree in electrical engineering from Georgia Tech.

Dr. Neil Zeghidour

Kyutai France

Abstract: In this talk, I will present Kyutai's work on Moshi, the first full-duplex conversation system ever released. Moshi is a speech-to-speech LLM which can provide natural conversations with human-like latency and emotions. I will unveil the technical details behind Moshi and explain how it can unlock a new generation of AI companions.

Bio: Neil is co-founder and Chief Modeling Officer of the Kyutai non-profit research lab. He was previously at Google DeepMind, where he started and led a team working on generative audio, with contributions including Google’s first text-to-music API, a voice preserving speech-to-speech translation system, and the first neural audio codec that outperforms general-purpose audio codecs. Before that, Neil spent three years at Facebook AI Research, working on automatic speech recognition and audio understanding. He graduated with a PhD in machine learning from Ecole Normale Supérieure (Paris), and holds an MSc in machine learning from Ecole Normale Supérieure (Saclay) and an MSc in quantitative finance from Université Paris Dauphine. In parallel with his research activities, Neil teaches speech processing technologies at the École Normale Supérieure (Saclay).

Prof. Philip Torr

Oxford UK

Abstract: Applications of Generative AI (Gen AI) are expected to revolutionize a number of different areas, ranging from science & medicine to education. The potential for these seismic changes has triggered a lively debate about the potential risks of the technology, and resulted in calls for tighter regulation, in particular from some of the major tech companies who are leading in AI development. This regulation is likely to put at risk the budding field of open-source generative AI. Using a three-stage framework for Gen AI development (near, mid and long-term), we analyze the risks and opportunities of open-source generative AI models with similar capabilities to the ones currently available (near to mid-term) and with greater capabilities (long-term). We argue that, in the near term, overall, the benefits of open-source Gen AI outweigh its risks. As such, we encourage the open sourcing of models, training and evaluation data, and provide a set of recommendations and best practices for managing risks associated with open-source generative AI. The longer term is harder to predict, and in the second half of the talk I outline some hypothetical scenarios in which open source may not be so acceptable. However all that can be said with certainty as that we must constantly review how the field unfolds.

Bio: Professor Philip Torr did his PhD (DPhil) at the Robotics Research Group of the University of Oxford under Professor David Murray of the Active Vision Group. He worked for another three years at Oxford as a research fellow, and still maintains close contact as visiting fellow there.

He left Oxford to work for six years as a research scientist for Microsoft Research, first in Redmond, USA, in the Vision Technology Group, then in Cambridge founding the vision side of the Machine Learning and Perception Group. He then became a Professor in Computer Vision and Machine Learning at Oxford Brookes University. In 2013, Philip returned to Oxford as full professor where he has established the Torr Vision group.

He won several awards including the Marr prize (the highest honour in vision) in 1998. He is a Royal Society Wolfson Research Merit Award Holder. Recently, together with members of his group, he has won several other awards including an honorary mention at the NIPS 2007 conference for the paper 'P. Kumar, V. Kolmorgorov, and P.H.S. Torr, An Analysis of Convex Relaxations for MAP Estimation', in NIPS 21, Neural Information Processing Conference, and (oral) Best Paper at Conference for 'O. Woodford, P.H.S. Torr, I. Reid, and A.W. Fitzgibbon, Global Stereo Reconstruction under Second Order Smoothness Priors', in Proceedings IEEE Conference of Computer Vision and Pattern Recognition, 2008 . More recently he has been awarded best science paper at BMVC 2010 and ECCV 2010.

He was involved in the algorithm design for Boujou released by 2D3. Boujou has won a clutch of industry awards, including Computer Graphics World Innovation Award, IABM Peter Wayne Award, and CATS Award for Innovation, and a technical EMMY. He then worked closely with this Oxford based company as well as other companies such as Sony on the Wonderbook project.

He has been involved in numerous spin-outs including FiveAI, Oxsight and Eigent.

Fabian Güra

BluOrion Limited UAE

Abstract: Today’s state-of-the art LLMs tend to be fairly uniform with respect to the number of parameters spent at every layer, and during pre-training all parameters are trained all at once.

At the same time, research and practice offers clear indication that not all parts of the model are equally important for final task performance, giving rise to parameter efficient fine-tuning and post-processing steps such as Low-Rank Adaptation (LoRA) and model pruning.

This talk will introduce some ideas and results on how to approach pre-training fundamentally from a bottom up perspective: How can we make static and dynamic changes to model architectures and training schedules in pretraining such that a given model performance target can be achieved at a lower compute and parameter budget? How can you best train only the parameters you really need in the end?

Bio: Fabian is a Principal AI engineer at BluOrion Dubai, an LLM research lab he helped establish in Dubai this June. Before that he led various teams in Google at the interface of Assistant, Search, and Speech. Most recently he acted as the lead for factuality in Bard, guiding the team into the Gemini era. His educational background lies in engineering and high performance computing working for various chairs at TU Munich and ETH Zurich. Using mathematics together with a deep understanding of hardware and software to improve LLM performance is his core research interest.

Dr. Nicolas Granatino

STEM AI UK

Abstract: The talk will outline how a robust open source commitment fosters entrepreneurship, attracts talent and accelerate leadership in a given area. It will draw from recent research,  historical (Sun, Netscape, and Red Hat) and more recently in AI (OpenAI, Mistral, kyutai, Black Forest Labs).  In particular, the talk will show how AI researchers’ involvement in open source offers direct economic benefits, such as facilitating access to venture capital. This trend has inspired venture capital strategies that capitalise on robust open-source ecosystems, particularly in the AI sector, as seen in investment approaches in both Europe and Abu Dhabi. Moreover, open source engagement encourages the creation of ventures that are both mission-driven and globally focused, contributing to a dynamic and inclusive entrepreneurial landscape.

Bio: Nicolas Grana,no is a seasoned investor and entrepreneur. Nicolas, began his venture career at LabMorgan in 2000 and later co-founded Andurance Ventures LLP in 2011, with a par,cular focus on the creator economy. Since the publica,on of Meta's FAIR team's LLaMA paper in February 2023, Nicolas has focused on AI within the European ecosystem inves,ng in Mistral AI, being an ac,ve Board member at Kyutai—Europe's largest private AI lab—and most recently inves,ng in H. He also co-founded StemAI.vc an AI accelerator out of Paris dedicated to early-stage AI startups making over 20 investments since incep,on in March 2024.

Dr. Belgacem Haba

Adeia Corporation USA

Abstract: The semiconductor industry finds itself in difficult situation today due to few hard limitations in the manufacturing, assembly, and implementation of its devices. Artificial Intelligence (AI) has led the way and become one the major factors in this phenomenon because of its need for high bandwidth data processing. This latter is coupled with the difficulty of continuing with the Moore’s law in addition to the higher cost of manufacturing and implementation. This talk will focus on the difficulties introduced by AI on the semiconductor industry and discuss a few specific cases being proposed to tackle challenges such as chip architecture, device manufacturing, power delivery, and thermal solutions.

Bio: Dr. Belgacem Haba was born in 1957 in El-Meghaier, Algeria. Dr. Haba joined Adeia (previously Tessera) in 1996 and is now its Senior Technical Fellow and Vice President. Today he is heading the path finding team in its electronic R&D division. His latest activities include the development of 3D technologies for the evolution of microelectronics. Dr. Haba was with Google data center platform division as senior staff and before that he co-founded SiliconPipe Inc. in 2002, a high-speed interconnects Start-up company based in Silicon Valley that got acquired by Samsung. He also managed the advanced packaging R&D activities at Rambus. From 1988 to 1996, he worked for both the NEC Central Research Laboratories in Tokyo Japan and the IBM Watson Research Center in New York on the applications of lasers in microelectronics. Throughout his carrier, he contributed to the development of new technologies and products. Few of these technologies found their ways in hundreds of billions of electronic components.

Dr. Haba Holds a Ph.D. in materials science and engineering in 1988 from Stanford University, California in the field of solar energy. He also obtained from the same university two master’s degrees in applied physics and materials science and engineering. He received a DES in physics from the university of Houari Boumediene for science and technology, Algeria in 1980. Dr. Haba holds over 600 issued U.S. patents, and over 1700 patents and patent applications worldwide. He is listed among the top 150 most prolific inventors worldwide.

Prof. Roberto Di Cosmos

Software Heritage France

Abstract: There is a strong interplay between software development and machine learning:

AI models are providing new tools to develop software, while the inclusion of large pubicly available codebases in training datasets helps improve large language models' reasoning abilities, well beyond coding tasks.

In the specific domain of source code the issue of transparency of the training dataset assumes a special weight in the broader debate around open versus closed models.

Software Heritage, launched by Inria and in partnership with UNESCO, has been building the largest archive of publicly available source code for nearly a decade, and provides today the Software Hash Identifier for the over 50 billion software artifacts it collected from over 300 million projects, ensuring availability, guaranteeing integrity and enabling traceability of all its contents. Because of the core values that inform its approach to open access and code preservation, it is naturally concerned by these challenges.

In this talk we will start from the principled stance on the use of the Software Heritage archive for training models, report on the lessons learned from the collaboration with the BigCode project that created StarCoder2, and then focus on the challenges, ethical considerations, and technical limitations that arise in the current approaches to use open codebases in AI, in particular when it comes to transparency, accountability, and resource efficiency.

These limitations underscore the need for a Code Commons: a dedicated initiative to expand Software Heritage into a central resource for transparency, quality, accountability, and sustainability in machine learning on code.

By promoting transparency and responsible stewardship, Software Heritage aims to help researchers, developers, and organizations navigate the challenges of AI in code-based applications. This talk invites all stakeholders to collaborate on this ambitious vision.

Bio: An alumnus of the Scuola Normale Superiore di Pisa , with a PhD in Computer Science from the University of Pisa , Roberto Di Cosmo was associate professor for almost a decade at Ecole Normale Supérieure in Paris. In 1999, he became a Computer Science full professor at University Paris Diderot, where he was head of doctoral studies for Computer Science from 2004 to 2009. President of the board of trustees and scientific advisory board of the IMDEA Software institute and chair of the Software chapter of the National Committee for Open Science in France , he is currently on leave at Inria .

His research activity spans theoretical computing, functional programming, parallel and distributed programming, the semantics of programming languages, type systems, rewriting and linear logic, and, more recently, the new scientific problems posed by the general adoption of Free Software, with a particular focus on static analysis of large software collections. He has published over 20 international journal articles and 50 international conference articles .

In 2008, he created and coordinated the European research project Mancoosi , which had a budget of 4.4Me and brought together 10 partners to improve the quality of package-based open source software systems.

Following the evolution of our society under the impact of IT with great interest, he is a long term Free Software advocate, contributing to its adoption since 1998 with the best-seller Hijacking the world , seminars, articles and software. He created in October 2007 the Free Software thematic group of Systematic , which helped fund over 50 Open Source research and development collaborative projects for a consolidated budget of over 200Me. From 2010 to 2018, he was director of IRILL , a research structure dedicated to Free and Open Source Software quality.

He created in 2015, and now directs Software Heritage , an initiative to build the universal archive of all the source code publicly available, in partnership with UNESCO .

Dr. Neha Sengupta

G42 UAE

Abstract: In this talk, we will discuss the Jais family of Arabic-English bilingual models, comparing those trained from scratch with those adapted from off-the-shelf English LLMs. The models will be compared along several dimensions, including compute requirements, resulting proficiency in Arabic and English, evaluation benchmarks, and capability for cross-lingual transfer. In particular, we will explore the pros and cons of training from scratch vs adapting an existing model when adding a language as different from the base language as Arabic is to English.

Bio: Neha Sengupta, Principal Applied Scientist, Inception:
Neha leads the NLP Applied science team at Inception, G42 in Abu Dhabi. She joined G42 in September of 2018 where she has created several NLP systems including Arabic centric LLMs, conversational agents, translation systems, and NLP based data analytics engines.
She completed her PhD at the Indian Institute of Technology (IIT), Delhi, India in 2019, focusing on efficient algorithms for large scale graphs.
Prior to PhD, she worked at IBM Research Labs in Delhi, India, specializing in smart energy grids and resource scheduling.

Dr. Jingwei Zuo

Technology Innovation Institute UAE

Abstract: In this talk, we present Falcon Mamba 7B, a new base large language model based on the novel Mamba architecture. Falcon Mamba 7B is trained on 5.8 trillion tokens with carefully selected data mixtures. As a pure Mamba-based model, Falcon Mamba 7B surpasses leading open-weight models based on Transformers, such as Mistral 7B, Llama3.1 8B, and Falcon2 11B. It is on par with Gemma 7B and outperforms models with different architecture designs, such as RecurrentGemma 9B and RWKV-v6 Finch 7B/14B. Currently, Falcon Mamba 7B is the best-performing Mamba model in the literature at this scale, surpassing both existing Mamba and hybrid Mamba-Transformer models, according to the Open LLM Leaderboard. Due to its architecture, Falcon Mamba 7B is significantly faster at inference and requires substantially less memory for long sequence generation. Despite recent studies suggesting that hybrid Mamba-Transformer models outperform pure architecture designs, we demonstrate that even the pure Mamba design can achieve similar, or even superior results compared to the Transformer and hybrid designs. We make the weights of our implementation of Falcon Mamba 7B publicly available on under a permissive license.

Bio: Jingwei Zuo is currently a Senior Researcher at the AI Center of the Technology Innovation Institute (TII). He and his team are working on the Falcon Foundation Models, with a focus on exploring next-generation Large Language Models (LLMs) that incorporate emerging architectural designs beyond or in combination with Transformers. Jingwei received his PhD in Computer Science from Université Paris-Saclay and was awarded the Plateau de Saclay Doctoral Prize for the best scientific production in Information and Communication Science and Technology (ICST) in 2022.

Dr. Armand Joulin

Google DeepMind France

Abstract: This talk will introduce the Gemma 2 models, a series of open language models that are built based on knowledge distillation, and achieve competitive performance across all pretraining and post-training benchmarks. This models are tailored to run on devices, especially mobile and laptop.

Bio: Armand is a Research Director at Google DeepMind, in charge of the open version of Gemini called Gemma. Prior to this, he was a Research Director at Meta in charge of EMEA, where he supervised several key open projects such as LLaMA, DINO or fasttext.

Farouk Elghamry

Ruliad UAE

Abstract: In this talk, I will introduce the Deepthought open-source “reasoning” models developed by Ruliad, a series of open language models that are aimed to unlock test-time compute for all. We will discuss the model’s capabilities; technical details behind the first version of Deepthough; and what it means for the next generation of models.

Bio: Farouk is founding engineer of Ruliad, a UAE-based AI and development company on a mission to enable self-sovereign general intelligence for government and enterprise.

Prof. Jie Tang

Tsinghua University China

Abstract: Large language models have substantially advanced the state of the art in various AI tasks, such as natural language understanding and text generation, and image processing, multimodal modeling. In this talk, we will first introduce the development of AI in the past decades, in particular from the angle of China. We will also talk about the opportunities, challenges, and risks of AGI in the future. In the second part of the talk, we will use ChatGLM, an alternative but open sourced model to ChatGPT, as an example to explain our understandings and insights derived during the implementation of the model.

Bio: Jie Tang is a WeBank Chair Professor of Computer Science at Tsinghua University. He is a Fellow of the ACM, a Fellow of AAAI, and a Fellow of IEEE. His interest is artificial general intelligence (AGI). His research received the SIGKDD Test-of-Time Award (10-year Best Paper). He also received the SIGKDD Service Award.

Dr. Samuel Rodriques

FutureHouse USA

Abstract: I will introduce FutureHouse, a new non-profit AI research lab in San Francisco focused on building an AI Scientist to automate research in biology and other complex sciences. I will discuss our initial results at FutureHouse, including PaperQA2, the first AI system to achieve superhuman accuracy at recalling and summarizing information from the scientific literature; LAB-Bench, a new benchmark to evaluate language agent performance on diverse tasks in biology research; and Aviary, a new reinforcement learning framework for training complex language agents. I will also discuss our future roadmap, the impact that AI will have on science in the coming years, and the importance of open source models in the development of an AI Scientist.

Bio: Sam Rodriques is Director and CEO of FutureHouse, a new non-profit AI-for-science research laboratory in San Francisco aimed at automating research in biology and other complex sciences. Prior to starting Future House, Sam ran the Applied Biotechnology Laboratory at the Francis Crick Institute and invented several new biotechnologies, including a new nanofabrication method, a new approach to sensing neural activity with probes in the bloodstream, and new ways to extract spatial and temporal information from RNA sequencing. Sam was also one of the original architects of the Focused Research Organization model, which provides an alternative structure to scale up basic science projects outside of academia. Sam has co-founded four other companies, and obtained his PhD in Physics at the Massachusetts Institute of Technology in 2019. He has received numerous national awards and fellowships to support his studies and recognize his research, including the 2022 Allen Distinguished Investigator Award, 2019 Stat Wunderkind award, and the Myhrvold and Havranek Family Charitable Fund Hertz Graduate Fellowship.

Dr. Alhussein Fawzi

Google DeepMind UK

Abstract: In the modern era of computing, algorithms for solving fundamental problems such as vehicle routing for last-mile delivery, and energy load balancing are used billions of times every day. However, paradoxically, such algorithms solving bottleneck problems were often designed well before the advent of modern computation, and often date back to decades. This is because discovering new tailored algorithms is incredibly difficult. In this talk, I will describe how to leverage foundation models to discover new interpretable and trustworthy algorithms, with the potential to improve the efficiency of bottleneck systems.

Bio: Alhussein Fawzi is co-founder and CEO of a stealth startup, and until recently was a research scientist at Google DeepMind. He was awarded the MIT Innovators Under 35 award in 2023 for his work on discovering new algorithms with machine learning. He has worked on AI for Science, with a particular interest in using Machine Learning to unlock new results in Computer Science, and Algorithms. He recently published two papers in Nature magazine (2022 and 2023), where he used Machine Learning and Large Language Models to discover new algorithms for fundamental computational tasks, and new mathematical results. He has published more than 30 papers in major scientific journals and conferences.