Diego Calanzone

I study Artificial Intelligence Systems at the University of Trento.
My research interests include generative models, multi-task/multi-modal learning and emergence. I believe (1) knowledge stems from interaction as well, (2) language is not sufficient, (3) machines shall have a human nature as base. I’m particularly keen on probability theory.

In this web space I share my research in a broad sense, that is the pursuit of understanding about life, society and intelligence (my main field of studies) artificially and naturally. I love blogs seeking for rationality and wisdom [1][2], as well as learning about different ethnicities and cultures (big fan of National Geographic).

news

Mar 5, 2024	Our paper “Towards Logically Consistent Language Models via Probabilistic Reasoning” has been accepted at the ICLR 2024 Workshop on Reliable and Responsible Foundation Models!
Feb 5, 2024	I’m going to talk about “Acquiring Complex Concepts with Comparative Learning” in a poster session at the 2024 CORE Project Workshop, Universitat Pompeu Fabra, Barcelona!
Jun 8, 2023	Our paper “An open source perspective on AI and alignment with the EU AI Act” has been accepted to the AI4Safety workshop @ IJCAI 2023!
Apr 21, 2023	Good news: I will take part in the Cornell, Max Planck, Maryland Research School 2023 and the Mediterranean Machine Learning Summer School 2023. See you there! :)
Jan 8, 2023	My updated post on diffusion models from scratch is online here!

Selected readings

Multi-modal

A Generalist Agent

Reed et al. 2022

Abs

Inspired by progress in large-scale language modeling, we apply a similar approach towards building a single generalist agent beyond the realm of text outputs. The agent, which we refer to as Gato, works as a multi-modal, multi-task, multi-embodiment generalist policy. The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens. In this report we describe the model and the data, and document the current capabilities of Gato.
Multi-modal

Improving Multimodal Interactive Agents with Reinforcement Learning from Human Feedback

Abramson et al. 2022

Abs

Josh Abramson, Arun Ahuja, Federico Carnevale, Petko Georgiev, Alex Goldin, Alden Hung, Jessica Landon, Jirka Lhotka, Timothy Lillicrap, Alistair Muldal, George Powell, Adam Santoro, Guy Scully, Sanjana Srivastava, Tamara von Glehn, Greg Wayne, Nathaniel Wong, Chen Yan, Rui Zhu
Generative DL

Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise

Bansal, Goldstein et al. 2022

Abs

Standard diffusion models involve an image transform – adding Gaussian noise – and an image restoration operator that inverts this degradation. We observe that the generative behavior of diffusion models is not strongly dependent on the choice of image degradation, and in fact an entire family of generative models can be constructed by varying this choice. Even when using completely deterministic degradations (e.g., blur, masking, and more), the training and test-time update rules that underlie diffusion models can be easily generalized to create generative models. The success of these fully deterministic models calls into question the community’s understanding of diffusion models, which relies on noise in either gradient Langevin dynamics or variational inference, and paves the way for generalized diffusion models that invert arbitrary processes.