projects

Acquiring Complex Concepts with Comparative Learning

Learning to compose visual primitives with logical functions. Different techniques are tested, i.e. task hyper-networks and modular deep learning. Workshop paper presented at the 2024 CORE Workshop at Universitat Pompeu Fabra.

[paper]

Semantic comprehension in large language models and analogies with the human brain.

A seminar paper on mapping computational representations of language to brain activity, a comparison between representational spaces and questions on the intrinsic understanding of concepts.

[paper]

Evolution in Social Dilemma Games, Social Pressure and Emergent Behaviours

Learning collaborative and competitive strategies with multi-agent reinforcement learning and evolutionary algorithms in survival games.

[code] [paper]

Referring Expression Comprehension as Scene Graph Grounding

Efficient object recognition with linguistic references using Graph Neural Networks and CLIP.

[code]

Stable Point-E, 3D reconstruction from synthetic views

Expanding 3D reconstruction with diffusion models to multiple views with models for view synthesis.

[code] [paper]

DeepSWE, flood prediction with deep learning

Implementation of video prediction models to learn shallow water equations for floodings prediction from aeral views.

[code] [paper]