Karim Abdel Sadek

I am a first-year MSc student in Artificial Intelligence at the University of Amsterdam. I am broadly interested in Reinforcement Learning, Logic, (Algorithmic) Game Theory, and AI Alignment. I am currently working on goal misgeneralization and specification misgeneralization in RL with Usman Anwar at the Krueger AI Safety Lab, University of Cambridge. My prior research work includes learning theory and online learning, and more broadly theoretical computer science applied to ML.

Before my MSc, I graduated with a BSc in Mathematics and Computer Science for AI at Bocconi University. I was advised by Prof. Marek Elias, working on learning theory and theory of algorithms. The paper has been accepted at ICLR 2024. During my BSc, I spent the Spring ‘23 semester at Georgia Tech, supported with a full-ride scolarship.

I am always open and excited to collaborate on projects. If you are interested in working together, have a quick chat, or anything else, please reach out! You can e-mail me at karim dot abdel dot sadek at student dot uva dot nl

news

May 15, 2024	I am joining the Krueger AI Safety Lab as a Research Intern this summer in Cambridge! I will be supported by the ERA Fellowship.
Feb 15, 2024	I started to work on Goal Misgeneralization in RL with Usman Anwar at Krueger AI Safety Lab at the University of Cambridge.
Jan 15, 2024	My very first paper, ‘𝐀𝐥𝐠𝐨𝐫𝐢𝐭𝐡𝐦𝐬 𝐟𝐨𝐫 𝐂𝐚𝐜𝐡𝐢𝐧𝐠 𝐚𝐧𝐝 𝐌𝐓𝐒 𝐰𝐢𝐭𝐡 𝐫𝐞𝐝𝐮𝐜𝐞𝐝 𝐧𝐮𝐦𝐛𝐞𝐫 𝐨𝐟 𝐩𝐫𝐞𝐝𝐢𝐜𝐭𝐢𝐨𝐧𝐬’, has been accepted at ICLR with scores 8,8,8,8!
Sep 01, 2023	I started my MSc in Artificial Intelligence at the University of Amsterdam

selected publications

Algorithms for Caching and MTS with reduced number of predictions

Karim Abdel Sadek , and Marek Elias

International Conference on Learning Representations (ICLR), 2024

Abs PDF

ML-augmented algorithms utilize predictions to achieve performance beyond their worst-case bounds. Producing these predictions might be a costly operation – this motivated Im et al. (2022) to introduce the study of algorithms which use predictions parsimoniously. We design parsimonious algorithms for caching and MTS with action predictions, proposed by Antoniadis et al. (2023), focusing on the parameters of consistency (performance with perfect predictions) and smoothness (dependence of their performance on the prediction error). Our algorithm for caching is 1-consistent, robust, and its smoothness deteriorates with the decreasing number of available predictions. We propose an algorithm for general MTS whose consistency and smoothness both scale linearly with the decreasing number of predictions. Without the restriction on the number of available predictions, both algorithms match the earlier guarantees achieved by Antoniadis et al. (2023).
’Explaining RL decisions with trajectories’: A reproducibility Study

Karim Abdel Sadek , Matteo Nulli , Joan Velja , and 1 more author

Transactions on Machine Learning Research (TMLR), 2024

Abs PDF

This work investigates the reproducibility of the paper " Explaining RL decisions with trajectories “ by Deshmukh et al. (2023). The original paper introduces a novel approach in explainable reinforcement learning based on the attribution decisions of an agent to specific clusters of trajectories encountered during training. We verify the main claims from the paper, which state that (i) removing trajectories induces a lower initial state value, (ii) clusters present high-level behaviours, (iii) distant trajectories influence the decision of an agent, and (iv) humans correctly identify the attributed trajectories to the decision of the agent. We recover the environments used by the authors based on the partial original code they provided for one of the environments. While we confirm that (i), (ii), and (iii) partially hold, we extend on the largely qualitative experiments from the authors by introducing a quantitative metric to further support (iii), and new experiments and visual results for (i). Moreover, we investigate the use of different clustering algorithms and encoder architectures to further support (ii). We could not support (iv), given the limited extent of the original experiments. We conclude that, while some of the claims can be supported, further investigations and experiments could be of interest. We recognize the novelty of the work from the authors and hope that our work paves the way for clearer and more transparent approaches.