Carview!

CARVIEW

MOTORHOMES

Select Language

HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 last-modified: Wed, 15 Oct 2025 04:00:46 GMT access-control-allow-origin: * etag: W/"68ef1c6e-6cf9" expires: Mon, 29 Dec 2025 19:12:47 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: 7469:21D6A4:94D9BE:A6CBD0:6952D050 accept-ranges: bytes age: 0 date: Mon, 29 Dec 2025 19:02:47 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210093-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767034967.072504,VS0,VE240 vary: Accept-Encoding x-fastly-request-id: 7f8ec708b9fa72e5b1510fb7e68efb44d66e9229 content-length: 7012 Tahseen Rabbani

Tahseen Rabbani

I am an AI/ML postdoc at the University of Chicago, co-hosted by Ce Zhang and Tian Li. I'm primarily interested in efficiency, distributed learning and privacy. I'm also interested in numerical methods, generative watermarking, optimization, and machine translation.

News

2025
- I sat on the CFAgentic@ICML'25 Panel with Niloofar Mireshghallah, William Lindskog-Münzing, and Chi Wang to discuss the following timely question, "Will the future of AI agents be collaborative and federated?"
- I presented "Mitigating Unintended Memorization with LoRA in Federated Learning for LLMs" (joint with EPFL) as an oral at both the MemFM and CFAgentic workshops @ ICML 2025.
- July: After an exciting year at Yale working on applied ML, I am now returning to efficient and privacy-preserving ML and joining the University of Chicago as a Postdoctoral Scholar! I am very fortunate to be co-hosted by Tian Li and Ce Zhang.
- January: With Dr. Hartley's departure from Yale, I have since rotated over to Dr. Hoon Cho's lab! I am continuing my work in low-resource, private, distributed ML, but now for proteomics and drugs discovery.
- "HashEvict: A Pre-Attention KV Cache Eviction Strategy using LSH" is now on arxiv. We're pushing the Pareto frontier on speed and output quality for KV compression.
2024
- "Llama-3-Meditron: An Open-Source Suite of Medical LLMs" is accepted as an oral at the AAAI'25 Workshop AI2ASE.
- Co-hosted our watermarking workshop at NeurIPS'24. Catch the event here!
- Presented our distributed learning platform, "DISCO," at GDHF'24 in Nairobi, Kenya. This is a joint work with EPFL.
- Two works, "Shrinking the Size of Deep Extreme Multi-Label Classification" and "LSH-E Tells You What To Discard" are accepted to the NeurIPS'24 Workshop on Machine Learning and Compression.
- Our competition proposal "Erasing the Invisible: A Stress-Test Challenge for Image Watermarks" is accepted by NeurIPS'24.
- "WAVES: Benchmarking the Robustness of Image Watermarks" is accepted to ICML'24. This is a state-of-the-art benchmarking suite for image watermarks. We propose several powerful, novel attacks.
- I am defending in Spring 2024. Afterwards (2024-2025), I am heading to Yale LiGHT for a postdoc under Dr. Mary-Anne Hartley. I will be working on LLM compression and FL for healthcare analytics. Reach out if you'd like to collaborate!
- Selected as a 2024 RSA Security Scholar for my work in privacy-preserving ML.
2023
- Paper accepted to NeurIPS'23, "Large-Scale Distributed Learning via Private On-Device Locality-Sensitive Hashing." This work was also presented at the ICLR 2023 Workshop on Sparsity in Neural Networks.
- Presented a pre-print at SLowDNN'23 in Dubai, UAE, "A Framework for Representation and Fast Evaluation of Multilinear Operations in Convolutional TNNs."
2022
- Paper accepted to ICLR'23, "SWIFT: Rapid Decentralized Federated Learning via Wait-Free Model Communication." This work also received an abstract selection for the 2023 Qualcomm Innovation Fellowship and a spotlight at FL-NeurIPS'22.
- Paper accepted to NeurIPS'22, "SketchGNN: Scalable Graph Neural Networks," and two papers at the GL Frontiers 2022 workshop, the aforementioned "SketchGNN" and "Calibrated Dataset Condensation for Faster Hyperparameter Search."
- Oral presentation for a pre-print, "Comfetch: Federated Learning of Large Networks on Memory-Constrained Clients via Sketching" given at FL-AAAI-22.
- Paper accepted to MSML'22, "Practical and Fast Momentum-Based Power Methods."
2021
- Nominated as a 2022 Apple Scholar in AI/ML by the University of Maryland, College Park.
- Awarded an NSF COMBINE Fellowship for 2021-2022.

Published Papers

Benchmarking the Robustness of Image Watermarks

B. An*, M. Ding*, T. Rabbani* et al., 2024

We systematically reveal weaknesses in modern image-based watermarking protocols, including those of a generative variety. Check out our benchmark and toolkit at wavesbench.github.io/.

Large-Scale Distributed Learning via Private On-Device LSH

T. Rabbani*, M. Bornstein*, F. Huang. NeurIPS, 2023.

Using a new family of hash functions, we develop one of the first private, personalized, and memory-efficient on-device LSH frameworks for training recommender DNNs on extreme multi-label datasets.

SWIFT: Rapid Decentralized Federated Learning via Wait-Free Model Communication

M. Bornstein, T. Rabbani*, AS Bedhi, F. Huang. ICLR, 2023.

We enable wait-free model training for peer-to-peer FL using model caching. Our algorithm provable convergence at a SotA rate and empirically significantly speeds up global model convergence.

Sketch-GNN: Scalable Graph Neural Networks with Sublinear Training Complexity

M. Ding, T. Rabbani, B. An, EZ Wang, F. Huang. NeurIPS, 2022.

This paper proposes a sketch-based algorithm whose training time and memory grow sublinearly with respect to graph size by training GNNs atop a few compact sketches of graph adjacency and node embeddings.

Constructions of difference sets in nonabelian 2-groups

A. Applebaum, J. Clikeman, J. Davis, J. Dillon, J. Jedwab, T. Rabbani, K. Smith, W. Yolland. Algebra & Number Theory, 2023.

We determine that all groups of order 256 not excluded by the two classical nonexistence criteria contain a difference set, resolving a 25 year old question posed by John Dillon.

Practical and Fast Momentum-Based Power Methods

T. Rabbani*, A. Jain, A. Rajkumar, F. Huang. PMLR, 2022.

We provide a pair of novel momentum-based power methods, DMPower and a streaming variant, DMStream. In contrast with prior art, these accelerated methods do not depend on spectral knowledge.

Fast GPU Convolution for CP-Decomposed Tensorial Neural Networks

T. Rabbani*, A. Reustle*, F. Huang. IntelliSys, 2022.

We present a GPU algorithm for performing convolution with decomposed tensor products. We experimentally find up to 4.85x faster execution times than cuDNN for some tensors.

Nonabelian Orthogonal Building Sets

T. Rabbani*, K. Smith. Proceedings of the 14th International Conference on Finite Fields and their Applications, 2022.

We examine recent construction techniques of Hadamard difference sets in 2-groups and an extension of orthogonal building sets to nonabelian groups.

Unique minimal forcing sets and forced representation of integers by quadratic forms

T. Rabbani*. Rose-Hulman Undergraduate Mathematics Journal, 2016.

We use Bhargava’s theory of escalators to establish several infinite familes of positive integers, interpreted as singletons in N, without unique minimal forcing sets in T.

Improving the error-correcting code used in 3-G communication

T. Rabbani*. SIURO, 2015.

In 2011, Samsung Electronics Co. filed a complaint against Apple Inc. for alleged infringement of patents described in US 7706348, including a [30,10,10] code. We give a construction of an even better [30,10,11] non-cyclic code, which is distinct from the conventional BCH construction.

Workshop Papers

Mitigating Unintended Memorization with LoRA in Federated Learning for LLMs

T. Bossy*, J. Vignoud*, T. Rabbani, M. Jaggi, JRT Pastoriza. The Impact of Memorization on Trustworthy Foundation Models, 2025.

Fast Evaluation of Multilinear Operations in Convolutional Tensorial Neural Networks

T. Rabbani*, J. Su*, X. Liu, D. Chan, G. Sangston, F. Huang. Third Workshop on Seeking Low‑Dimensionality in Deep Neural Networks, 2023.

Faster Hyperparameter Search for GNNs via Calibrated Dataset Condensation

M. Ding*, Y. Xu, T. Rabbani, X. Liu, T. Ranadive, TC Tuan, F. Huang. New Frontiers in Graph Learning, 2023.

Comfetch: Federated Learning of Large Networks on Constrained Clients via Sketching

T. Rabbani*, B. Feng*, M. Bornstein, K. Sang, Y. Yang, A. Rajkumar, A. Varshney, F. Huang. International Workshop on Trustable, Verifiable and Auditable Federated Learning in Conjunction with AAAI, 2022.