Biography
Georges-André Silber is a researcher specializing in program compilation and optimization, with particular expertise in the analysis and transformation of nested loops. His work focuses in particular on polyhedral representations of loop nests, used for advanced optimizations such as merging, *tiling*, or *strip mining*, as well as on the trade-off between accuracy and efficiency in static analysis for production compilers. He has contributed to the development of tools such as GRAPHITE (for GCC) and WRaP-IT, in collaboration with teams from INRIA and the University of Paris-Sud, as well as to the analysis of induction variables via abstract structures such as *trees of recurrences*. His research also includes the manipulation of Fortran programs (via the Nestor library) and the study of mixed dependencies for loop distribution with control structures.
Publication(s)
-
2006
GRAPHITE: Polyhedral analyses and optimizations for GCC
-
2005
Induction variable analysis with delayed abstractions DOI : 10.1007/11587514_15
-
2000
Temporary arrays for distribution of loops with control dependences DOI : 10.1007/3-540-44520-x_47
-
1999
The Nestor library: A tool for implementing fortran source to source transformations DOI : 10.1007/bfb0100626
-
1998
Loop parallelization algorithms: From parallelism extraction to code generation DOI : 10.1016/S0167-8191(98)00020-9
Teaching
Natural Language Processing
Introduction This course explores how natural language processing (NLP) applications work, which today rely primarily on machine learning approaches. The goal for an engineer is, in particular, to understand the technical workings and scientific foundations of large language models (LLMs) such as ChatGPT and to be able to create such a model using libraries like PyTorch. The course is divided into lectures and hands-on exercises in roughly equal proportions. Indicative Content NLP Data - characters, words, texts, tokens - from Baudot to OpenAI’s tiktoken Formal Languages and NLP - Computer Science and formal languages, Chomsky hierarchy - Rational, algebraic, and contextual languages; Turing machines NLP Techniques - Tokenization, stemming, lemmatization, POS, NER, ... - Sentiment analysis, text mining, similarity, ... Deep Learning and NLP - Language models, LLMs (autoregressive models, Transformer) - BERT, LLaMA, GPT, ... - Energy-based models (EBM) NLP Software - Solr (Lucene), NLTK - PyTorch
Information Systems Management (ISM) track
The MSI track complements the core and specialized coursework to provide students with a thorough understanding of these technologies: various technical topics are covered depending on the academic year. Examples include cryptography, network security, scripting languages for rapid development, source code management tools, computer language compilation, the use and configuration of virtual machines, distributed systems, and more. Finally, an introduction to management issues helps students understand the challenges of effectively implementing these technologies in organizations’ information systems. Examples of projects within the track, either group or individual: development of a “MineCoin” blockchain; analysis of hacking traces using machine learning; development of a 2D game with a custom engine; web-based recommendation applications (finance, humor); mobile online gaming app; network attacks via packet interception and modification; cryptographic instant messaging; cryptographic remote backup system; truly secure “cloud” data sharing; filtering algorithms on graphics processing units (GPUs); automatic analysis of data schemas Examples of elective topics from recent years: classification using speech recognition and semantic analysis (Deezer); data analysis and revenue forecasting (Datadog, NYC); integration and optimization of detour suggestions (Blablacar); development of a customer behavior tracking system (Pandacraft); development of visual effects (Smart & Soft); Android application security audit (Solucom); insurance pricing using geographic data (AXA); map enrichment of radar views (THALES); high availability for an in-memory database (Quartet FS); security analysis of the OpenFlow protocol (ANSSI); information system for Autolib (Polyconseil); analysis of “big data” on the airline market (AMADEUS); design of targets for radar testing (THALES); compilation and optimization of image processing applications (KALRAY); optimization of a distributed data caching system (AMADEUS, Germany); design of a web-based customer portal (SopraGroup / LexisNexis) optimization of video content delivery (Orange Lab) development of graphics-based games (Silicon Studio Corp, Japan) decision-making architecture models (Solucom) automatic construction of user profiles (Technicolor, USA) migration of server architectures to the cloud (Samsung, South Korea) operating system optimized for supercomputers (CEA) mobile video application (TangoMe) detection of banking fraud (Oresys) financial risk calculations (Société Générale) oil production monitoring application (Accenture / Total); Google Earth interface extension (Google, USA); financial application instrumentation (CALYON, Japan)
Introduction to Computer Science
PhD supervision
- 2023 Towards more energy-efficient software with new cost models for optimizing compilers SOLCOURT Antoine
- 2023 Static cost models of compilers optimizing for improving the energy performance of embedded software DESTAL Benjamin