Danilo de Goede

PhD student of Artificial Intelligence

University of Amsterdam, QUVA Lab

Biography

I’m a PhD student at the QUVA lab at the University of Amsterdam supervised by Yuki Asano and Cees Snoek. Additionally, I am enrolled in the ELLIS PhD program in collaboration with Qualcomm. My research focuses on improving the adaptation efficiency in foundation models. More broadly, I am interested in understanding and improving the underlying learning mechanics of representation learning methods.

Previously, I completed my Master’s degree in Artificial Intelligence at the University of Amsterdam, where I mainly worked on computer vision and causal representation learning. I also hold a Bachelor’s degree in Computer Science from the University of Amsterdam.

Interests

Learning Mechanics
Image Tokenization
Self-Supervised Learning
Foundation Models

Education

MSc in Artificial Intelligence, 2023
University of Amsterdam
BSc in Computer Science, 2021
University of Amsterdam

News

Oct 2026: Our preprint “Semantically Steerable Image Tokenization” is out on arXiv.
Sep 2023: I have started my PhD on Adaptable Foundation Models at the University of Amsterdam under the supervision of Cees Snoek and Yuki Asano.
Apr 2023: I gave a tutorial on Causal Identifiability from Temporal Intervened Sequences as a part of the Causality and Causal Representation Learning module of the Deep Learning 2 course at the University of Amsterdam.
Jan 2023: I gave an invited guest lecture “MLRC 2021 - Our Experience” at the introductory lecture of the Fairness in AI course of the University of Amsterdam.
Aug 2022: Our paper “C-3PO: Towards Rotation Equivariant Feature Detection and Description” has been accepted to the VIPriors workshop at ECCV 2022.
Jul 2022: Our poster on “Quantifying CLIP’s Ability to Perform Cross-Modal Grounding using Attention-Model Explainability” received the best poster award at the XAI course of the University of Amsterdam.
Apr 2022: Our paper on the Reproducibility Study of “Counterfactual Generative Networks” received the Best Paper Award of MLRC 2021.
Jan 2022: Our paper “The Cost of Reinforcement Learning for Game Engines: The AZ-Hive Case-study” has been accepted to ICPE 2022.
Jun 2021: Our presentation “The Cost of AlphaZero: the Hive Case” received the best presentation award at CompSys-2021.

Recent Publications

Semantically Steerable Image Tokenization

We introduce GIST, an adaptive image tokenizer whose token allocation is jointly controlled by pixel-level and semantic target errors, so that heavy compression can discard pixel detail while preserving what the image depicts.

Danilo de Goede, Mohammad Mahdi Derakhshani, Yuki M. Asano, Cees G. M. Snoek

C-3PO: Towards Rotation Equivariant Feature Detection and Description

We propose C-3PO, a family of feature detection-and-description models based on steerable group convolutions.

Piyush Bagad, Floor Eijkelboom, Mark Fokkema, Danilo de Goede, Paul Hilders, Miltiadis Kofinas

Reproducibility Study of “Counterfactual Generative Networks”

We reproduce the paper “Counterfactual Generative Networks” by Axel Sauer and Andreas Geiger to identify which parts of the experimental results supporting their claims can be reproduced, and at what cost in terms of resources (e.g., computational cost, development effort, and communication with the authors).

Piyush Bagad, Jesse Maas, Paul Hilders, Danilo de Goede

The Cost of Reinforcement Learning for Game Engines: The AZ-Hive Case-study

We analyse the cost and efficiency of the AlphaZero approach when building a new game engine. Thus, we present our experience building AZ-Hive, an AlphaZero-based playing engine for the game of Hive.

Danilo de Goede, Duncan Kampert, Ana Lucia Varbanescu