Differences

This shows you the differences between two versions of the page.

--- start [2025/10/10 14:42] – [Large Language Models, Graphs and Applications (LLGA)] cxypolop
+++ start [2025/10/28 16:09] (current) – [Internships and PhD proposals] director-trai
@@ Line 2: / Line 2: @@
-** Programme of the Master of Science and Technology (MScT) of Ecole Polytechnique **
+**Programme of the Master of Science and Technology (MScT) of École polytechnique**
+[[https://programmes.polytechnique.edu/en/master/all-msct-specializations/trustworthy-and-responsible-ai-trai|Official website]]
-[[https://programmes.polytechnique.edu/master/programmes/trustworthy-and-responsible-ai-trai|Link to the official page on the Ecole Polytechnique website. ]]
 ==== Motivation ====
-The **LLGA Master’s Program** is a two-year, fully English-taught degree with a strong international outlook. Designed for ambitious students, the program equips graduates with both a solid theoretical foundation and hands-on industry expertise in **Generative AI**. With a special emphasis on **Large Language Models (LLMs)** and **graph-structured data**, students gain the skills to harness cutting-edge technologies and apply them across diverse fields—from healthcare and biology to finance, social networks, and beyond.
+Artificial intelligence (AI) is at the core of modern information systems upon which we increasingly rely to solve complex tasks such as selecting job candidates, automating daily tasks, analyzing medical data, and controlling critical systems. This has led to growing societal impact and increasing academic and industrial interest, even more so with the emergence of generative AI. Due to its pervasive use, improving the //reliability//, //safety//, and //robustness// of AI systems as well as accounting for their ethical and legal aspects have become a necessity. Indeed, there is growing evidence that left untethered, AI leads to biased, prejudiced solutions prone to tampering, hallucinations, and unsafe behaviors. This shows how far we still are from being able to build //trustworthy AI systems//.
+The TRAI master’s program prepares its graduates to face these growing challenges and become experts in this transformative field, with skills to advance both theory and application.
 ==== Objectives ====
-In recent years, Generative AI has seen explosive growth, with Large Language Models (LLMs) reshaping how society and the economy interact with the digital world. These technologies enable tasks that were unimaginable just a few years ago—from machine translation and code generation to creative content generation and advanced reasoning —at unprecedented scale. The objective of the LLGA Master’s Program is to prepare students to become experts in this transformative field, with the skills to advance both theory and application. Equally important are graph-based data and models, which capture the structure of complex relationships in social networks, molecular structures, or logistics chains. Advances in Graph Neural Networks (GNNs) and Graph Representation Learning are unlocking insights and applications previously out of reach. Combining LLMs with graph-based methods enables multimodal generative AI, opening new opportunities in healthcare, finance, recommendation systems, and beyond.
+The goal of the TRAI program is to provide students with the theoretical and practical tools needed to build //trust// and //reliability// into AI systems. At its core lie the seven key requirements for Trustworthy AI established by the high-level expert group on AI of the European Comission:
-The goal of the LLGA program is to provide students with a solid theoretical foundation in machine and deep learning, alongside hands-on experience with state-of-the-art generative techniques in the context of LLMs and graph generation. Through projects, case studies, and industrial applications, graduates acquire rare and highly sought-after expertise, positioning them to innovate and lead in the next generation of AI technologies.
+  - human agency and oversight
+  - technical robustness and safety
+  - privacy and data governance
+  - transparency
+  - diversity, non-discrimination and fairness
+  - environmental and societal well-being
+  - accountability
-Courses are given by professors from Ecole Polytechnique and partner institutes and companies.
+To achieve these specifications, students will develop a solid mathematical foundation to understand the limitations of and develop guarantees for machine learning, deep learning, and generative models. They will also be brought to think about the role of these techniques in fighting biases and desinformation, both at the societal level and within companies. By means of projects and industrial case studies, students will acquire hands-on experience with state-of-the-art trustworthy AI methods, positioning them to innovate and lead the development and deployment of the next generation of AI technologies in critical applications.
-All courses will be held in English. Both French and foreign students are welcome.
+Courses are taught by professors from École polytechnique, associated institutes, and industrial partners.
+**All courses are taught in English. Students from all nationalities are welcome.**
-==== Curriculum Description  ====
+==== Curriculum description ====
-The LLGA Master is a two-year program that combines a rigorous foundation with advanced specialization in Generative AI, focusing on both Large Language Models (LLMs) and graph-structured data.
+The program begins with a week of refresher courses in mathematics (statistical analysis, machine learning) and computer science (programming, basics of algorithms) aimed to complement the student academic background. This is followed by two training periods of 2.5 months, consisting of general courses on topics such as
-During the first year (M1), students build a strong academic basis through courses shared with engineering and other master’s programs at École Polytechnique. This year develops solid skills in mathematics, machine learning, and deep learning, while also introducing key domains such as Natural Language Processing and Graph Representation Learning. By the end of M1, students are equipped with the theoretical and methodological tools essential to pursue advanced AI studies.
+  * Optimization and statistics for AI
+  * Deep learning, generative models, and reinforcement learning
+  * Large language models (LLMs)
-The second year (M2) is dedicated to specialization and hands-on learning. Bespoke courses focus on the latest industrial and scientific advances, including DevOps and LLM Engineering Principles, Data Engineering for LLMs, Graph Generative AI with Applications in Bio and Medicine, Small-Scale and Specialized LLMs, AI Strategy, Ethics, and Socioeconomic Challenges, and Generative AI for Entrepreneurship. Students engage in projects and case studies that connect theory with real-world applications, gaining practical experience with state-of-the-art techniques.
+as well as specific courses targeting trustworthiness and reliability requirements, such as
-This unique structure—broad foundations followed by tailored specialization—ensures graduates acquire both depth and versatility, preparing them to become leaders in the rapidly evolving field of Generative AI.
+  * Privacy, bias, and fairness for AI
+  * Security, robustness, and verification for AI
+  * Explainability, sustainability, and frugality
-==== Student Applications ====
+A weekly seminar is held to encourage discussions of ethical issues as well as entrepreneurship and innovation, while transverse projects provide practical, real-world experience to the students.
-Students are required to have a suitable background in mathematics and computer science. In practice, this means knowledge of linear algebra, statistics, as well as Python programming, algorithms and data structures. Some exposure to natural language processing and graph theory is desirable, but not required. Candidates with Bachelor degrees that are not in Mathematics or Computer Science are also admissible if the required knowledge domains are covered in their degree programs.
+A more detailed description of the curriculum is available [[curriculum|here]].
-==== Industrial & Institutional Partnership / Cercle des partenaires ====
-Students will have highly suitable profiles for data science or AI roles in industry. While the LLM aspect of the program allows students to apply to companies in most digital companies, the graphs and applications aspect of the program prepares the students particularly well for roles in the pharma, logistics and communication industries.  Graduates of the LLGA Master’s Program are prepared for careers as AI/ML Engineers, Data Scientists, NLP Specialists, Graph AI Experts, and AI Research Scientists, as well as strategic roles such as AI Product Managers, Solution Architect.  Opportunities span industries including tech, healthcare, finance, consulting, social media, e-commerce, and public institutions.
+Students also take part in two five- to six-month [[#internships_and_phd_proposals|research projects or internships]], conducted either in a public or private research lab.
+==== Student applications ====
-==== Internships and PhD proposals ====
+Students are expected to have to have a strong background in mathematics and computer science, i.e., knowledge of linear algebra, basic optimization, and statistics as well as algorithms, data structures, and programming (Python and others). Candidates with Bachelor degrees that are not in Mathematics, Engineering, or Computer Science are also admissible as long as the required knowledge are covered in their degree programs.
-The program includes an M1 as well as an M2 internship of several months each. There will furthermore be regular interaction with professionals in industry in academia in the weekly M2 seminar series on “ethical issues, law and novel applications of AI”. In addition, many of the taught courses in the MScT program have project-based assessments. The students will therefore graduate from this program with ample practical experience and a profound insight into the different real-world applications of their academic knowledge. We also expect important exchanges with the generative AI industry in terms of focused internships, invited keynotes and meetups. We expect that the placement of our graduates in the relevant industry will be natural and swift.
+Please refer to the [[https://programmes.polytechnique.edu/en/master/admissions-msct/application-deadlines-procedure|official documentation]] for details on the application process.
-Since the program teaches cutting-edge methodology in a relevant field of research, pursuing a PhD after this program is perfectly possible.
+==== Industrial & institutional partnerships (Cercle des partenaires) ====
+The TRAI master’s program equips graduates with specialized skills that are particularly valuable for industries where reliability and transparency are crucial. Graduates can go on to take key roles in organizations dealing with critical applications and/or prioritizing ethical and interpretable AI, including energy, transport, healthcare, ecological transition, and finance. Graduates of the TRAI master’s program are prepared for careers as AI/ML engineers, data scientists, and AI research scientists, as well as strategic roles such as AI ethics specialist, system auditor, and solution architect.
-==== Contacts ====
-The academic co-directors of the MScT are:
+==== Internships and PhD proposals ====
-  * Johannes Lutzeyer, Computer Science, Ecole Polytechnique [[johannes.lutzeyer@polytechnique.edu]]
-  * Aymeric Dieuleveut, Applied Mathematics, Ecole Polytechnique [[aymeric.dieuleveut@polytechnique.edu]]
-  * Michalis Vazirgiannis, Computer Science, Ecole Polytechnique [[michalis.vazirgiannis@polytechnique.edu]]
-  * Ye Zhu, Computer Science, Ecole Polytechnique [[ye.zhu.lix@polytechnique.edu]]
-  * Rémi Flamary, Applied Mathematics, Ecole Polytechnique  [[remi.flamary@polytechnique.edu]]
-  * Luiz Chamon, Applied Mathematics, Ecole Polytechnique [[luiz.chamon@polytechnique.edu]]
+The TRAI program includes two five- to six-months internships periods in their first (M1) and second (M2) year. Students are free to find their own positions or may choose to apply to one of the internship proposals collected [[internships|here]] (regularly updated).
+This program covers cutting-edge methodology in a relevant and timely domain of research. Pursuing a PhD after concluding this master’s is therefore perfectly possible, be it purely academic or in collaboration with the industry (e.g., by means of a program known in France as [[https://www.campusfrance.org/en/what-involved-Doctorate-France|CIFRE]]). More information in French [[https://www.enseignementsup-recherche.gouv.fr/fr/les-cifre-46510|here]].
+==== Contacts ====
+The academic co-directors of the MScT are:
+  * Luiz Chamon, Applied Mathematics (CMAP / DepMAP), École polytechnique ([[https://luizchamon.com/|website]])
+  * Karim Lounici, Computer Science (LIX / DIX), École polytechnique ([[https://klounici.github.io/|website]])
+  * Jesse Read, Computer Science (LIX / DIX), École polytechnique ([[https://jmread.github.io/|website]])
+  * Sonia Vanier, Computer Science (LIX / DIX), École polytechnique ([[https://www.lix.polytechnique.fr/~vanier/|website]])
-Please don't hesitate to contact us!
+Please don’t hesitate to contact us by email at (<dir.msct.trai@polytechnique.fr>)!