Carlos Eduardo Rosar Kos Lassance
Résumé: english Google Scholar
Currently
Member of Technical Staff at Cohere, working on the embeddings team developping new solutions for information retrieval needs.
Professional Experience
01/2024-
- Member of Technical Staff at Cohere.
11/2020-12-2023
- Research Scientist at Naver Labs Europe, on the Neural Search project. Mainly working on information retrieval with sparse neural retrievers:
- Multilingual neural retrieval: Won both tracks @ WSDM 2023 CUP - MIRACL , Training SPLADE from scratch in Arabic, Japanese and Russian, TREC-NeuCLIR 2022 participation (article coming soon)
- Consolidation of SPLADE as an effective and efficient sparse neural retriever: Splade++, Efficiency study for SPLADE, TREC-DL2021 participation, TREC-DL2022 participation, SPLADE repository ,Reproducibility via Pyserini, Training retrieval models from scratch @ ECIR 23 ,Learning adapters for SPLADE (accepted @ ECIR23 coming soon)
- Improving the efficiency of learned representations for retrieval: TLDR, ColBERT pruning, CCSA
10/2017-10/2020
- PHD Student at IMT Atlantique
PHD Student at IMT Atlantique, with the theme “Graphs for Deep Learning Latent Representations”, under the supervision of Vincent Gripon and Michel Jezequel. The Ph.D. program in the intersection of Graph Signal Processing (GSP) and Machine Learning. The main goal is to better understand the intermediate representations generated by Deep Neural Networks (DNN) and to use that understanding/knowledge to improve them. During this three years period, I’ve interned at the Mila research lab in Montreal for one year and was a visiting scholar in the universities of Rochester and Adelaide. Projects were developped mostly with Python, Pytorch and Pandas. The most relevant projects are:
- Matching CNN without priors about data. Published at IEEE DSW 2018
- Increasing the robustness of DNNs. Colaboration with USC. Arxiv only
- Evaluating robustness of DNNs. Colaboration with Mila and USC. Published at IEEE DSW 2019
- Neural network compression with graph distillation. Colaboration with Mila and USC, Published at ICASSP 2020 Video presentation available
- Graph topology inference benchmarks for machine learning. Colaboration with the University of Rochester. To be presented at MLSP2020.
- Improving visual based localization using graphs. Colaboration with University of Adelaide.
10/2018-10/2019
- Internship at Mila under the supervision of Jian Tang on the subject of graphs and deep neural networks.
01/2017-07/2017
- Intern at the LEARN laboratory at PUC-Rio
Worked in the development of the first version of the BTG Pactual Robo Advisor. Micro-service using Flask, deployed via Amazon cloud and integrated with Amazon Redshift database.
09/2015-09/2016
- R&D Engineer in neural networks at Télécom Bretagne in NEUCOD team
R&D Engineer at Télécom Bretagne in NEUCOD team under the supervision of Claude Berrou, working mostly with artificial and clique-based neural networks. My work at the laboratory helped the team to publish 2 papers in internacional conferences. Development of clique-based associative memories in CUDA and neural networks for handwritten digit identification with Theano.
03/2015-09/2015
- Intern at INRIA/Centrale Supéléc in CIDRE team under the supervision of Sébastien Gambs and Cristina Onete.
Studied distance-bounding protocols and implemented a proof of concept in Android that got published in CARDIS 2015 and was the main part of my MSc thesis.
01/2012-08/2013
- Teaching assistant at PUC-Rio for the databases 101 course.
08/2011-08/2013
- Internship at Labbio/BioBD a Bio-informatics and databases research lab at PUC-RIO.
Development of a Java program to migrate the database hosting the DPVAT (brazilian national vehicular assurance) to a new schema and DBMS. Development of PHP websites for:
- Administration of a summer camp
- Administration of social programs
- University course registration
- Dashboard for a twitter crawler
- Dashboard for the university’s private bus
Teaching Experience
IMT-Atlantique
- Substitute teacher (6 hours) for TC131E(Pyrat: Algorithms and Programming), 7.5 hours of courses and organization of the PyRat Challenge for ELU502(Deep Learning), and full participation (21 hours) in ELU616(Artificial Intelligence) at IMT-Atlantique
MOOC
- Creation of jupyter notebook exercises and integration with Vocareum for the Advanced Algorithmics and Graph Theory with Python course on edX
PUC-Rio
- 3 tutorials for the INF1395(Machine Learning) course at PUC-Rio, concerning Theano, Pandas and Perceptron. Helped create the pandas exercises for the 2017 version of the INF1026(Applied Computing) course
Education
2017-20
PHD Student at IMT Atlantique, with the theme “Intricate Learning and Storing in Deep Neural Networks”, under the supervision of Vincent Gripon and Michel Jezequel. Ph.D. Defense schelduled for September 2020.
2014-15
MSc in Informatic Research from Télécom Bretagne: Track: Mobile and Communicating Object-Based Systems. Thesis: Distance bounding protocols on smartphones
2010-17
BSc/MEng in Computer Engineering: Double-degree program between Télécom Bretagne and PUC-RIO (One year break working as R&D Engineer between 2015 and 2016.)
Publications (outdated, please check scholar for recent publications)
2020
C. Lassance, M. Bontonou, G.B. Hacene, V. Gripon, J. Tang and A. Ortega, “Deep Geometric Knowledge Distillation with Graphs
“, Barcelona, Spain, 2020. [Paper] [Presentation]
2019
C. Lassance, V. Gripon, J. Tang and A. Ortega, “Structural Robustness for Deep Learning Architectures,” 2019 IEEE Data Science Workshop (DSW), Minneapolis, USA, 2019. [Paper] [Presentation]
Lassance, C. E. R. K., Gripon, V., & Ortega, A. (2018). Laplacian Networks: Bounding Indicator Function Smoothness for Neural Network Robustness. Graph Signal Processing Workshop, June 2019, Minneapolis, United States. [Paper] [Poster]
Bontonou, M.\^, Lassance, C.^, Hacene, G. B., Gripon, V., Tang, J., & Ortega, A. (2019). Introducing Graph Smoothness Loss for Training Deep Learning Architectures. 2019 IEEE Data Science Workshop (DSW), Minneapolis, USA. (^Equal contribution) [Paper] [Presentation]
Bontonou, M., Lassance, C., Gripon, V., & Farrugia, N. (2019, September). Comparing linear structure-based and data-driven latent spatial representations for sequence prediction. In Wavelets and Sparsity XVIII (Vol. 11138, p. 111380Z). International Society for Optics and Photonics. [Paper]
Bontonou, M., Lassance, C., Vialatte, J. C., & Gripon, V. (2019). A Unified Deep Learning Formalism For Processing Graph Signals. Graph Signal Processing Workshop, June 2019, Minneapolis, USA. Short Paper
Hacene, G. B., Lassance, C., Gripon, V., Courbariaux, M., & Bengio, Y. (2019). Attention Based Pruning for Shift Networks. arXiv preprint arXiv:1905.12300. [Paper]
2018
Lassance, Carlos Eduardo Rosar Kos, Vincent Gripon and Antonio Ortega__. “Predicting Under and Overfitting in Deep Neural Networks using Graph Smoothness” Graph Signal Processing Workshop, June 2018, Lausanne, Switzerland. [Paper] [Presentation]
Lassance, C. E. R. K., Vialatte, J. C., & Gripon, V. (2018, June). Matching Convolutional Neural Networks without Priors about Data. In 2018 IEEE Data Science Workshop (DSW) (pp. 234-238). [Paper] [Poster]
Grelier, N., Lassance, C. E. R. K., Dupraz, E., & Gripon, V. (2018, November). Graph-Projected Signal Processing. In 2018 IEEE Global Conference on Signal and Information Processing (GlobalSIP) (pp. 763-767). IEEE. [Paper] [Poster]
2017
Marques, M. R. S., Hacene, G. B., Lassance, C. E. R. K., & Horrein, P. H. (2017, July). Large-Scale Memory of Sequences Using Binary Sparse Neural Networks on GPU. In 2017 International Conference on High Performance Computing & Simulation (HPCS) (pp. 553-559).
2016
Tigreat, P., Rosar Kos Lassance, C., Jiang, X., Gripon, V., Berrou, C.__ Assembly output codes for learning neural networks. 9th International Symposium on Turbo Codes and Iterative Information Processing, Sept 2016, Brest, France
2015
Gambs, S., Lassance, C. E. R. K., & Onete, C. (2015, November). The Not-so-Distant Future: Distance-Bounding Protocols on Smartphones. In International Conference on Smart Card Research and Advanced Applications (pp. 209-224). Springer, Cham. (Authors in alphabetical order)
Technical skills
- Python (Incluiding frameworks such as Transformers and Pytorch)
- Java
- Machine learning
- LaTeX
- C and C++
- Git / SVN
- UNIX
- Databases (PostgreSQL and MySQL)
- Bash
- HTML and CSS
- PHP
- Javascript
Awards
BRAFITEC scolarship (MSc fees)