Publications

ImmunoFoundation: A Multimodal Foundation Model for Immunogenicity Prediction and Peptide Optimization
João Felipe Rocha
Hiren Madhu
Jenny Yongjia Liu
Apurva Mishra
Chen Liu
Rishabh Anand
Rex Ying
Smita Krishnaswamy
Peptide immunogenicity, whether a peptide presented by an MHC molecule elicits a T-cell response, is central to designing vaccines, cancer i… (see more)mmunotherapy, and therapeutic proteins. Existing tools rely on a single modality, such as peptide sequences or peptide-MHC interactions, and often ignore the T-cell response that depends on the TCR-peptide-MHC complex (TCR-pMHC) and its three-dimensional structure. The scarcity of labeled TCR-pMHC data with known structures makes it difficult to build a model that captures how all components of the TCR-pMHC contribute to immunogenicity. However, a foundation model of TCR-pMHCs can learn transferable representations across components, which can be adapted to immunogenicity, binding, and TCR specificity tasks, even with limited labeled data. We introduce **ImmunoFoundation**, a self-supervised multimodal backbone for protein-complex representation, fine-tuned for peptide--MHC immunogenicity. The model couples an ESM-2 sequence encoder with a graph transformer over structure, fused via cross-modal attention. Pretraining follows a curriculum that progressively introduces structural inductive bias. **ImmunoFoundation** ourperforms prior multimodal class-I predictors on cancer neoepitope and infectious-disease tasks.
2026-05-27
FM4LS @ International Conference on Machine Learning (poster)
openreview.net
openreview.net
Leveraging Routing Dynamics in Mixture-of-Experts Models for Efficient Language Adaptation
Aditi Khandelwal
Marius Mosbach
Verna Dankers
Siva Reddy
Golnoosh Farnadi
Mixture-of-Experts (MoE) models are widely used to scale language models, yet their expert routing behavior and adaptation in a multilingual… (see more) setting remain underexplored. In this work, we study multilingual routing dynamics during continual pre-training of an English-centric MoE model on a multilingual corpus, analyzing how expert usage varies across languages. We find that continual multilingual pre-training leads to diffused, language-agnostic routing in early and middle layers, with language specialization primarily emerging in the final layers. We also show that token-level vocabulary overlap between languages plays an important role in how languages are routed. Motivated by these findings, we propose a parameter-efficient adaptation strategy that updates language-specific and shared experts in the final MoE layers. Experiments on MultiBLiMP and Belebele show that our method achieves a strong performance-efficiency trade-off, attaining competitive performance relative to fine-tuning complete final layers, while updating less than 2% of the parameters. Overall, our findings provide insights into where and how language specialization emerges in MoEs during continual pre-training and provide practical insights for low-resource multilingual adaptation. Our code is available at https://github.com/aditi184/moe-routing-adaptation.
2026-05-27
arXiv (preprint)
doi.org
arxiv.org
Rethinking Literature Search Evaluation: Deep Research Helps, and Human Citation Lists Are Not a Ground Truth
Gaurav Sahu
Laurent Charlin
Christopher Pal
We study large-scale literature search from two complementary angles: improving the retrieval pipeline, and stress-testing the human referen… (see more)ce list as an evaluation target. First, we implement a Deep Research pipeline that processes the full query paper and expands the retrieved results breadth-first along their bibliographies, and show that it substantially outperforms vanilla API-only search, raising recall on RollingEval-Jun25 (a 250-paper literature-search benchmark) from below 20% to above 80%. Second, we use a neutral LLM-as-a-judge to determine if human references are sound ground truth for the task. We find significant limitations: only 51% of human citations are judged moderately relevant or higher, against 86--88% for the strongest AI-based re-rankers. We study this gap on the OpenAlex co-authorship graph, finding that humans are 2.5x more likely than the best AI re-rankers to cite a direct collaborator. Together, our results argue against single-axis literature-search evaluation: recall, topical-relevance scoring, ranked-list diversity, and a co-authorship-distance diagnostic each measure complementary properties of citation quality and should be reported jointly.
2026-05-27
arXiv (preprint)
doi.org
arxiv.org
UniSafe: Modality-Agnostic Hateful Content Detection via Shared-Space Projection
Siam Shibly Antar
Syem Shibly Ador
Steven H. H. Ding
Benjamin C. M. Fung
2026-05-27
ACM Web Conference (published)
doi.org
BlabberSeg: Semantic Perception for Reliable Open-Vocabulary UAV Safe Landing
Haechan Mark Bong
Giovanni Beltrame
Reliable robot autonomy requires semantic perception that remains both informative and fast enough for closed-loop safety decisions. We pres… (see more)ent BlabberSeg, an optimized CLIPSeg-based open-vocabulary segmentation pipeline for UAV emergency landing. The method targets semantic reliability under edge constraints by reusing prompt, positional, and image features and deploying floating-point 16 ONNX (TensorRT) inference. In a DOVESEI-based safe-landing workflow, BlabberSeg reaches 16.78Hz on Jetson Orin AGX (64GB), a 927.41% speed increase over the original CLIPSeg (1.81Hz), with limited degradation in segmentation agreement (2.1% relative area difference) and mIoU (9%). At the task level, safe-landing success is preserved (76/100, matching baseline) while mission time is substantially reduced. These results support semantic open-vocabulary perception as a practical component for reliable autonomous landing.
2026-05-26
SRRA @ IEEE International Conference on Robotics and Automation (poster)
openreview.net
openreview.net
Dr-CiK: A Testbed for Foresight-Driven Agents
Yihong Tang
Andrew Williams
Arjun Ashok
Vincent Zhihao Zheng
Lijun Sun
Alexandre Drouin
Issam H. Laradji
Étienne Marcotte
Valentina Zantedeschi
Time series forecasting in real-world settings often depends not only on historical observations, but also on external context that must be … (see more)actively discovered from noisy, heterogeneous information sources. Yet existing context-aided forecasting benchmarks typically assume that the supporting context is already provided, leaving open whether agents can identify it on their own. Therefore, we introduce Dr-CiK, a benchmark for evaluating whether agents can retrieve forecasting-relevant supporting context from a document corpus, filter out distractors, distill the retrieved context into forecast-useful evidence, and generate forecasts supported by that evidence. Through context ablations and evaluations of state-of-the-art deep research and forecasting methods paired together, we show that high-quality context substantially improves forecasting performance in Dr-CiK. However, most existing DR agents recover only a small fraction of the ground-truth supporting evidence (usually <5%), are frequently misled by distractors (>80% distractor citations), and can cause forecasters to perform worse with retrieved context than without context. Our results motivate research on foresight-driven agents that search for the right context to predict the future.
2026-05-26
arXiv (preprint)
doi.org
arxiv.org
Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions
Katie Everett
Elliot Paquette
Existing theory of momentum assumes that gradients arrive at every parameter at a roughly constant rate, an assumption violated in practice … (see more)by heavy-tailed data distributions and modern architectures. We theoretically analyze the dynamics of two tractable models of momentum under sparse updates: a least squares model with sparse inputs and a logistic regression model with a rare class. Both admit exact closed-form second-moment dynamics whose high-dimensional limits we characterize across three scaling exponents for sparsity, batch size, and momentum decay. The phase structure on both problems is governed by the ratio of two intrinsic timescales: a momentum retention timescale (how many active updates the buffer survives) and a learning timescale (how many active updates it takes to reduce the squared error). When learning is much slower than retention, the limit matches SGD; when learning is faster, the system is unstable; where the timescales coincide, we recover classical heavy-ball dynamics. The oscillatory dynamics occur at different momentum values for different token sparsity, creating a spectral conflict for global momentum across token frequencies.
2026-05-26
arXiv (preprint)
doi.org
arxiv.org
FedEHR-Gen: Federated Synthetic Time-Series EHR Generation via Latent Space Alignment and Distribution-Aware Aggregation
Jun Bai
Ziyang Song
Yue Li
Synthetic Electronic Health Record (EHR) generation provides a promising avenue for data augmentation and cross-hospital modeling in privacy… (see more)-constrained healthcare settings. However, most existing EHR generative models are centralized and require pooling data across hospitals, which is often infeasible when real-world data sharing is restricted. While federated EHR generation offers a natural solution, direct federated modeling often collapses or diverges due to the high dimensionality, sparsity, and cross-hospital heterogeneity of EHR data. In this work, we propose FedEHR-Gen, the first federated framework for synthetic time-series EHR generation across distributed hospitals. FedEHR-Gen uses a two-stage learning paradigm. First, we introduce a federated autoencoder that projects high-dimensional and sparse EHR features onto a compact latent space. To ensure semantic consistency across hospitals, we develop a layer-wise matching aggregation mechanism that aligns local encoders into a unified global latent space. Second, operating on this aligned latent space, we train a federated temporal conditional variational autoencoder (TCVAE) with distribution-aware aggregation, enabling stable temporal generative modeling under severe cross-hospital heterogeneity. Extensive experiments on the eICU and MIMIC-III datasets demonstrate that FedEHR-Gen achieves generation fidelity, downstream utility, and privacy risk comparable to centralized training, while consistently outperforming the standard federated baseline.
2026-05-26
arXiv (preprint)
doi.org
arxiv.org
HoloQ-VLA: Uniform W4A4 Quantization of Vision-Language-Action Models
Xinyu Wang
Mingze Li
Sicheng Lyu
Dongxiu Liu
Kaicheng Yang
Ziyu Zhao
Yufei Cui
Xiao-Wen Chang
Peng Lu
Vision-Language-Action (VLA) models unify perception, reasoning, and control in a single policy, but their multi-billion-parameter backbones… (see more) and diffusion-based action heads make on-device deployment prohibitively expensive. Low-bit post-training quantization (PTQ) is the natural remedy, yet the diffusion action head that emits continuous control signals is highly sensitive to it: a few weight and activation outliers are enough to destabilize the head, so prior work leaves it at full precision or falls back to mixed-precision schemes, and uniformly quantizing the whole model to low bit-width remains an open challenge. We present HoloQ-VLA, the first training-free PTQ framework that compresses both the language backbone and the entire diffusion action head to uniform W4A4 precision without mixed-precision allocation. Instead of trading weight quality against activation quality, HoloQ-VLA targets the two outlier sources with complementary transforms: a weight-adapted rotation composed with an activation-dispersing Hadamard transform, together with per-step scaling that absorbs the dynamic-range drift exhibited by the action head across denoising steps. On LIBERO, HoloQ-VLA compresses Pi-0.5 and GR00T-N1.5 to W4A4 with 98.0% and 87.8% task success rates, matching or exceeding their FP16 references of 97.1% and 87.0%, while reducing the static memory footprint by 74.2%. Real-world manipulation experiments further demonstrate that HoloQ-VLA maintains smooth and accurate control across diverse real-world scenarios.
2026-05-26
arXiv (preprint)
doi.org
arxiv.org
Online Sketched Newton-Raphson
Jean-Luc Lupien
Yuen-Man Pun
Youssef Diouane
Iman Shames
Antoine Lesage-Landry
In online convex optimization (OCO), a decision-maker is confronted with an unknown environment and seeks to play an optimal sequence of dec… (see more)isions on a short time-scale using only past information. Recent advances in second-order OCO methods have demonstrated tighter regret bounds and improved empirical performance over traditional first-order methods. However, this performance comes at a cost: a matrix inversion is now required, which scales with the cube of the size of the problem. In this work, we propose sketching to mitigate this limitation. Specifically, we present the online sketched Newton-Raphson method (OSNR) which preserves the tight regret bounds obtained with second-order methods while presenting a strict computational improvement in terms of complexity. We discuss three application scenarios of OSNR: online root finding, unconstrained OCO, and time-varying equality-constrained OCO, and present their respective regret and a constraint violation bound for the latter. In all three applications, OSNR achieves sublinear dynamic regret bounds. For the equality-constrained case, the extension OSNR with equality constraints OSNR-EC is shown to yield sublinear cumulative constraint violation. Finally, we illustrate the performance of OSNR and OSNR-EC on two numerical examples, viz., online position tracking and optimal power flow, and observe that OSNR and OSNR-EC exhibit high performance even at low sampling rates.
2026-05-26
arXiv (preprint)
doi.org
arxiv.org
Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation
Vitoria Barin Pacela
Shruti Joshi
Isabela Camacho
Simon Lacoste-Julien
David Klindt
The linear representation hypothesis states that neural network activations encode high-level concepts as linear mixtures. However, under su… (see more)perposition, this encoding is a projection from a higher-dimensional concept space into a lower-dimensional activation space, and a linear decision boundary in the concept space need not remain linear after projection. In this setting, classical sparse coding methods with per-sample iterative inference leverage compressed sensing guarantees to recover latent factors. Sparse autoencoders (SAEs), on the other hand, amortise sparse inference into a fixed encoder, introducing a systematic gap. We show this amortisation gap persists across training set sizes, latent dimensions, and sparsity levels, causing SAEs to fail under out-of-distribution (OOD) compositional shifts. Through controlled experiments that decompose the failure, we identify dictionary learning -- not the inference procedure -- as the binding constraint: SAE-learned dictionaries point in substantially wrong directions, and replacing the encoder with per-sample FISTA on the same dictionary does not close the gap. An oracle baseline proves the problem is solvable with a good dictionary at all scales tested. Our results reframe the SAE failure as a dictionary learning challenge, not an amortisation problem, and point to scalable dictionary learning as the key open problem for sparse inference under superposition.
2026-05-26
CompLearn @ International Conference on Machine Learning (poster)
doi.org
openreview.net
Coordinating speech and body during conversation: an at-home study with 4-year-olds
Chantal-Valerie Lee
Guillaume Dumas
Ahmed Jérôme Romain
Simone Falk
2026-05-25
International Conference on Speech Prosody (published)
doi.org