Mathieu Desponds's repositories
Robust-journey-planner-with-CFF-data
Proposed optimal routes by considering the delays and the correspondence probability where an adapted Raptor algorithm was developed.
Language-Models-Fundations
We build two language model (an LSTM and a Transformer model). Then we build models on the downstream task of **Sentence Paraphrasing** by fine-tuning an Encoder-Decoder architecture with attention and a Transformer.
Decoder-in-Natural-Language-Generation
Implementation of greedy and beam search, as well as top-p and top-k. Analyze how varying specific parameters of decoding and sampling algorithms can qualitatively affect the generation and interpretation of NLG evaluation metrics.
Transfer-Learning-and-Data-Augmentation
1. Fine-tune DistilBERT on NLI and dentify the some salient or toxic features that the model learnt. 2. Sample annotations techniques and production of silver label using (EDA and Back Translation).
Information-extraction-in-official-documents-using-LLMs
Assessed MistralAI-7B capabilities for document information extraction while ensuring client confidentiality, using In-Context Learning, Chain-of-Thought, and LoRa fine-tuning. Develop cost-effective strategies for deploying LLMs in production environments
NLP552-educational-chat-1
First part of the final project of CS522 Natural Language Processing
NLP552-educational-chat-3
This is the 3rd part of the final project of NLP-552. The goal of the project is to use T5 to create a chat-bot for EPFL students using RLHF.
ML433-Censored-Tweet-Analysis
Analysis of censored tweets. Undestanding the topics that are censored in different countries using different NLP techniques
ML433-Higgs-Boson-Detection
A data science project, that goes from exploratory data analysis to feature processing and engineering. The goal of the project is to implement machine learning techniques on the data, evaluate their models, generate predictions, and report the findings.
Data-visualization-NBA-data-story
Data story of a NBA season using D3. Use maps and charts to make a data visualization.
UCN-over-faulty-network
Implementation of Uniform Causal Network over faulty network from scratch using UDP and standard Java library.
Evaluating-Machine-Translation-as-preprocessing-step
Try to anwer the question whether for a new language, we should train a new model or we can translate to english, use english SOTA models and translate back if necessary