Hamid's Personal Homepage
Latest News
Introduction
I am an AI research scientist at Meta Super Intelligence Lab (MSL), and part of the LLaMa post-training team. My research focus spans various direction including post-training, reasoning, and reward modeling in LLMs, studying representation of DNNs and phenomenon such as space folding, fundamental aspects of RL such as information-theoric principles in offline RL, generalization in transformer architectures for in-context retrieval, and multimodal vlm proactive agents. At Meta, I led human preference alignment at MSL, improving LLaMA’s performance on user alignment, unverifiable capabilities, reasoning, and response quality. I also led research on in-context generative retrieval for Meta's ranking, and vlm proactive multimodal agents for Meta's smart wearables. Before joining Meta, I was a Postdoc at the Institute for Machine Learning, at the Johannes Kepler University of Linz. In my research, I am interested in alignment, post training, reasoning, and large-scale reinforcement learning.
Selected publications
Here is a list of my selected publications.
- Dhruvesh Patel, Hamid Eghbal-zadeh, Nitin Kamra, Michael Louis Iuzzolino, Unnat Jain, Ruta Desai Pretrained Language Models as Visual Planners for Human Assistance, In Proceedings of International Conference on Computer Vision (ICCV), 2023.
- Marius-Constantin Dinu, Markus Holzleitner, Maximilian Beck, Hoan Duc Nguyen, Andrea Huber, Hamid Eghbal-zadeh, Bernhard A. Moser, Sergei Pereverzyev, Sepp Hochreiter, Werner Zellinger Addressing Parameter Choice Issues in Unsupervised Domain Adaptation by Aggregation, In Proceedings of International Conference on Learning Representations (ICLR), 2022.
- Fabian Paischer, Thomas Adler, Vihang Patil, Angela Bitto-Nemling, Markus Holzleitner, Sebastian Lehner, Hamid Eghbal-zadeh, Sepp Hochreiter History Compression via Language Models in Reinforcement Learning, In Proceedings of International Conference on Machine Learning (ICML), 2022.
- Kajetan Schweighofer, Andreas Radler, Marius-Constantin Dinu, Markus Hofmarcher, Vihang Patil, Angela Bitto-Nemling, Hamid Eghbal-zadeh, Sepp Hochreiter A Dataset Perspective on Offline Reinforcement Learning, In Proceedings of Conference on Lifelong Learning Agents (CoLLAs), 2022.
- Martin Gauch, Maximilian Beck, Thomas Adler, Dmytro Kotsur, Stefan Fiel, Hamid Eghbal-zadeh, Johannes Brandstetter, Johannes Kofler, Markus Holzleitner, Werner Zellinger, Daniel Klotz, Sepp Hochreiter, Sebastian Lehner Few-Shot Learning by Dimensionality Reduction in Gradient Space, In Proceedings of Conference on Lifelong Learning Agents (CoLLAs), 2022.
- Christian Steinparz, Thomas Schmied, Fabian Paischer, Marius-Constantin Dinu, Vihang Patil, Angela Bitto-Nemling, Hamid Eghbal-zadeh, Sepp Hochreiter Reactive Exploration to Cope with Non-Stationarity in Lifelong Reinforcement Learning, In Proceedings of Conference on Lifelong Learning Agents (CoLLAs), 2022.
- Khaled Koutini, Jan Schlüter, Hamid Eghbal-zadeh, Gerhard Widmer Efficient training of audio transformers with patchout In Proceedings of Interspeech, 2022.
- W. Zellinger, N. Shepeleva, M.C. Dinu, H. Eghbal-zadeh, H.D. Nguyen, B. Nessler, S. Pereverzyev, B.A. Moser, The balancing principle for parameter choice in distance-regularized domain adaptation, In Proceedings of Advances in Neural Information Processing Systems (NeurIPS), 2021.
- K. Koutini, H. Eghbal-zadeh, G. Widmer, Receptive Field Regularization Techniques for Audio Classification and Tagging with Deep Convolutional Neural Networks, in IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP), vol. 29, pp. 1987-2000, 2021, doi: 10.1109/TASLP.2021.3082307. [paper][code]
- H. Eghbal-zadeh, F. Henkel, G. Widmer, Context-Adaptive Reinforcement Learning using Unsupervised Learning of Context Variables, In Proceedings of Machine Learning Research, NeurIPS 2020 Workshop on Pre-registration in Machine Learning, PMLR 148:236-254, 2021. [paper][blog post]
- H. Eghbal-zadeh, F. Henkel, G. Widmer, Learning to Infer Unseen Contexts in Causal Contextual Reinforcement Learning, SSL-RL Workshop, ICLR, 2021. [paper][environment]
- H. Eghbal-zadeh, F. Henkel, G. Widmer, Context-Adaptive Reinforcement Learning using Unsupervised Learning of Context Variables, Pre-registration Workshop, NeurIPS, 2020. [blog post]
- H. Eghbal-zadeh, K.Koutini, V. Haunschmid, P. Primus, M. Lewandowski, W. Zellinger, G. Widmer, Adversarial Robustness in Data Augmentation, Towards Trustworthy ML: Rethinking Security and Privacy for ML, ICLR Workshop, 2020. [talk]
- H. Eghbal-zadeh, Representation Learning and Inference from Signals and Sequences, PhD Thesis, 2019.
- H. Eghbal-zadeh, W. Zellinger, G. Widmer, Mixture Density Generative Adversarial Networks, In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019. [paper] [code]
- H. Eghbal-zadeh, M. Dorfer, G. Widmer, Deep Within-Class Covariance Analysis for Robust Audio Representation Learning, Advances in Neural Information Processing Systems, Interpretability and Robustness in Audio, Speech, and Language Workshop, 2018. [paper] [slides]
- H. Eghbal-zadeh, L. Fischer, N. Popitsch, F. Kromp, S. Taschner-Mandl, T. Gerber, E. Bozsaky, P. F. Ambros, I. M. Ambros, G. Widmer, B. A. Moser, DeepSNP: An End-to-End Deep Neural Network with Attention-Based Localization for Breakpoint Detection in Single-Nucleotide Polymorphism Array Genomic Data, in Journal of Computational Biology, 2018. [paper] [code]
Talks
2021
Title of the talk: How Much Is An Augmented Sample Worth?
Description: Teaser at the Pre-register workshop at NeurIPS2021.
Title of the talk: Learning Context Variables in Contextual Reinforcement Learning
Description: Talk at the 5th IPM Advanced School On Computing & Artificial Intelligence (2021).
Title of the talk: Foundations of Data Augmentation
Description: Talk at the Amirkabir Artificial Intelligence Summer Summit (AAISS) 2021.
2020
Title of the talk: Context-Adaptive Reinforcement Learning using Unsupervised Learning of Context Variables
Description: Teaser at the Pre-register workshop at NeurIPS2020.
Title of the talk: CP lecture serries
Description: Public lecture at the CP lecture serries.
Title of the talk: On the Inductive Biases in Data Augmentation and Adversarial Robustness.
Description: Talk at the Amirkabir Artificial Intelligence Summer Summit (AAISS) 2020.
Title of the talk: Adversarial Robustness in Data Augmentation.
Description: Talk at the Towards Trustworthy ML: Rethinking Security and Privacy for ML, ICLR 2020 Workshop.
2019
Title of the talk: Representation Learning under Incomplete and Imprecise Information Conditions.
Description: Talk at the Intelligent Autonomous Systems Group of the Computer Science Department of the Technische Universitaet Darmstadt.
2018
Title of the talk: Deep Within-Class Covariance Analysis for Robust Deep Audio Representation Learning.
Description: Talk at the Interpretability and Robustness in Audio, Speech and Language, NeurIPS 2018 Workshop.
Mentoring
If you are from an under represented group, and need help with ML research or similar topics, you can book a mentoring session with me.