📋 Tender Overview
Deep Cognition Learning For Language Generation
Unlocking The Future Of Language Understanding And Processingin Recent Years, Transformer-Based Deep-Learning Models Like Bert And Gpt-3 Have Dazzled Us With Their Natural Language Processing (Nlp) Prowess, Showcasing Remarkable Transfer And Few-Shot Learning Abilities. However, These Models Often Stumble When Pushed Beyond Their Comfort Zones, Lacking Out-Of-Domain Adaptability, Contextual Understanding, Calibration And Traceable Memory. Funded By The European Research Council, The Decollage Project Will Address These Challenges. Specifically, It Will Develop Utility-Guided Controlled Generation With Uncertainty Estimates. It Will Also Integrate Contextual Information Efficiently, And Craft Sparse Memory Models. Decollage Fosters Conscious Processing Through Attention Descriptive Representations And Enables Seamless Communication Among Modules And Agents. With Applications Spanning Machine Translation, Open Dialogue And Story Generation, Decollage Will Ensure Nlp’S Evolution Stays Ahead Of The Curve.
Funded Under
European Research Council (Erc)
Total Cost
€ 1 999 595,00
Eu Contribution
€ 1 999 595,00
Coordinated By
Instituto De Telecomunicacoes
Portugal
