Publications
Our teams aspire to make discoveries that impact everyone, and core to our approach is sharing our research and tools to fuel progress in the field.
Our teams aspire to make discoveries that impact everyone, and core to our approach is sharing our research and tools to fuel progress in the field.
Sort By
1 - 15 of 10822 publications
mmMUSE: An mmWave-based Motion-resilient Universal Speech Enhancement System
Chenming He
Yanyong Zhang
Kai Wang
Dequan Wang
Lingyu Wang
the Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT), ACM (2026) (to appear)
Preview abstract
Voice-based smart systems can greatly enhance user experiences by allowing higher-quality interactions through better voice perception. Speech enhancement can benefit such systems by isolating noise from speech. Recently, integrating millimeter-wave (mmWave) with audio for speech perception has gained increasing attention due to microphones' limitations in noisy environments. However, mmWave-based vocal extraction is severely affected by motion, which disperses vocal signals across ranges and introduces distortions. In this paper, we propose an mmWave-based motion-resilient universal speech enhancement system called mmMUSE, which fuses mmWave and audio signals. To mitigate motion interference, we develop a Doppler-based method for motion-robust vocal signal extraction. Moreover, by introducing the Vocal-Noise-Ratio metric to assess the prominence of vocal signals from mmWave, we achieve real-time voice activity detection that gains 3.81 dB of SISDR in noisy speeches. Additionally, we design a two-stage complex-valued network that includes an attention-based fusion network for cross-modal complementing and a time-frequency masking network for correcting amplitude and phase of speech to isolate noises.
Using mmWave and audio datasets from 46 participants, mmMUSE outperforms the state-of-the-art speech enhancement models, achieving an average SISDR improvement of 3.12 dB. Additionally, mmMUSE achieves SISDR improvements of 16.51 dB, 17.93 dB, 14.93 dB, and 18.95 dB in controlled environments involving intense noise, extensive motion, multiple speakers, and various obstructive materials, respectively. Finally, we evaluate mmMUSE in real-world scenarios including running, public spaces, and driving, maintaining a word error rate (WER) below 10%.
View details
Preview abstract
AI coding assistants are rapidly becoming integral to modern software development. A key challenge in this space is the continual need to migrate and modernize codebases in response to evolving software ecosystems. Traditionally, such migrations have relied on rule-based systems and human intervention. With the advent of powerful large language models (LLMs), AI-driven agentic frameworks offer a promising alternative—but their effectiveness remains underexplored. In this paper, we introduce FreshBrew, a novel benchmark for evaluating AI-based agentic frameworks on project-level Java migrations. We benchmark several such frameworks, powered by state-of-the-art LLMs, and compare their performance against established rule-based tools. Our evaluation of AI agents on this benchmark of 228 repositories shows that the top-performing model, Gemini 2.5 Flash, can successfully migrate 56.5% of projects to JDK 17. Our empirical analysis reveals novel insights into the critical strengths and limitations of current agentic approaches, offering actionable insights into their real-world applicability. By releasing FreshBrew publicly upon acceptance, we aim to facilitate rigorous, reproducible evaluation and catalyze progress in AI-driven codebase modernization.
View details
Preview abstract
For many practical applications of quantum computing, the slowest and most costly steps involve coherently accessing classical data. We help address this challenge by applying mass production techniques, which can sometimes allow us to perform operations many times in parallel for a cost that is comparable to a single execution[1-3]. We combine existing mass-production results with modern approaches for loading classical data using ``quantum read-only memory.'' We show that quantum mass production techniques offer no benefit when we consider a cost model that focuses purely on the number of non-Clifford gates. However, analyzing the constant factors in a more nuanced cost model, we find that it may be possible to obtain a reduction in cost of an order or magnitude or more for a variety reasonably-sized fault-tolerant quantum algorithms. We present several applications of quantum mass-production techniques beyond naive parallelization, including a strategy for reducing the cost of serial calls to the same data loading step.
View details
Differentially Private Insights into AI Use
Daogao Liu
Pritish Kamath
Alexander Knop
Adam Sealfon
Da Yu
Chiyuan Zhang
Conference on Language Modeling (COLM) 2025 (2025)
Preview abstract
We introduce Urania, a novel framework for generating insights about LLM chatbot interactions with rigorous differential privacy (DP) guarantees. The framework employs a private clustering mechanism and innovative keyword extraction methods, including frequency-based, TF-IDF-based, and LLM-guided approaches. By leveraging DP tools such as clustering, partition selection, and histogram-based summarization, Urania provides end-to-end privacy protection. Our evaluation assesses lexical and semantic content preservation, pair similarity, and LLM-based metrics, benchmarking against a non-private method inspired by CLIO (Tamkin et al., 2024). Moreover, we develop a simple empirical privacy evaluation that demonstrates the enhanced robustness of our DP pipeline. The results show the framework’s ability to extract meaningful conversational insights while maintaining stringent user privacy, effectively balancing data utility with privacy preservation.
View details
Toward Sensor-In-the-Loop LLM Agent: Benchmarks and Implications
Zhiwei Ren
Junbo Li
Minjia Zhang
Di Wang
Longfei Shangguan
SenSys 2025 - The 23rd ACM Conference on Embedded Networked Sensor Systems (2025)
Preview abstract
This paper advocates for sensor-informed personal agents that can take advantage of sensor hints on wearables to enhance the personal agent's response. We demonstrate that such a sensor-in-the-loop design paradigm can be easily integrated into existing LLM agents by building a prototype named WellMax based on existing well-developed techniques such as structured prompt tuning and few-shot prompting. The head-to-head comparison with a non-sensor-informed agent across five use scenarios demonstrates that this sensor-in-the-loop design can effectively improve users' needs and their overall experience. The deep-dive into agents' replies and participants' feedback further reveals that sensor-in-the-loop agents not only provide more contextually relevant responses but also exhibit a greater understanding of user priorities and situational nuances. We further conduct two case studies to examine the potential pitfalls and distill key insights from this sensor-in-the-loop agent. We believe this work sets the stage for more intelligent, empathetic, and effective interactions in future AI-driven personal assistants.
View details
Improved FPT Approximation Scheme and Approximate Kernel for Biclique-Free Max k-Weight SAT: Greedy Strikes Back
Theoretical Computer Science, 1028 (2025)
Preview abstract
In the Max k-Weight SAT (aka Max SAT with Cardinality Constraint) problem, we are given a CNF formula with n variables and m clauses together with a positive integer k. The goal is to find an assignment where at most k variables are set to one that satisfies as many constraints as possible. Recently, Jain et al. (SODA 2023) gave an FPT approximation scheme (FPT-AS) with running time 2^O((dk/ε)^d) * (n + m)^O(1) for Max k-Weight SAT when the incidence graph is K_{d,d}-free. They asked whether a polynomial-size approximate kernel exists. In this work, we answer this question positively by giving an (1 − ε)-approximate kernel with (dk/ε)^O(d) variables. This also implies an improved FPT-AS with running time (dk/ε)^O(dk) * (n+m)^O(1)-time algorithm for the problem. Our approximate kernel is based mainly on a couple of greedy strategies together with a sunflower lemma-style reduction rule.
View details
Synthetic Text Generation for Training Large Language Models (LLMs) via Gradient Matching
Dang Nguyen
Zeman Li
Meisam Razaviyayn
Baharan Mirzasoleiman
International Conference on Machine Learning (ICML) (2025)
Preview abstract
Synthetic data has the potential to improve the performance, training efficiency, and privacy of
real training examples. Nevertheless, existing approaches for synthetic text generation are mostly heuristics and cannot generate human-readable text without compromising the privacy of real data, or provide performance guarantees for training Large Language Models (LLMs). In this work, we propose the first theoretically rigorous approach for generating synthetic human-readable text that provides convergence, performance, and privacy guarantees for fine-tuning LLMs on a target task. To do so, we leverage Alternating Direction Method of Multipliers (ADMM) that iteratively optimizes the embeddings of synthetic examples to match the noisy gradient of the target training or validation data, and maps them to a sequence of text tokens with low perplexity. In doing so, the generated synthetic text guarantees convergence of the model to a close neighborhood of the solution obtained by fine-tuning on real data and preserves their privacy. Experiments on various classification tasks confirm the effectiveness of our proposed approach. Our code is available at https://github.com/BigML-CS-UCLA/GRADMM.
View details
Machine learning the effects of many quantum measurements
Wanda Hou
Samuel Garratt
Norhan Eassa
Yi-Zhuang You
Ehud Altman
arXiv (2025)
Preview abstract
Measurements are essential for the processing and protection of information in quantum computers. They can also induce long-range entanglement between unmeasured qubits. However, when post-measurement states depend on many non-deterministic measurement outcomes, there is a barrier to observing and using the entanglement induced by prior measurements. Here we demonstrate a new approach for detecting such measurement-induced entanglement. We create short-range entangled states of one- and two-dimensional arrays of qubits in a superconducting quantum processor, and aim to characterize the long-range entanglement induced between distant pairs of qubits when we measure all of the others. To do this we use unsupervised training of neural networks on observations to create computational models for post-measurement states and, by correlating these models with experimental data, we reveal measurement-induced entanglement. Our results additionally demonstrate a transition in the ability of a classical agent to accurately model the experimental data; this is closely related to a measurement-induced phase transition. We anticipate that our work can act as a basis for future experiments on quantum error correction and more general problems in quantum control.
View details
Applying multimodal AI to physiological waveforms improves genetic prediction of cardiovascular traits
Yuchen Zhou
Mahantesh I. Biradar
Jacqueline Shreibati
Dongbing Lai
Tae-Hwi Schwantes-An
Robert Luben
Zachary R. McCaw
Jorgen Engmann
Rui Providencia
Amand Floriaan Schmidt
Patricia B. Munroe
Howard Yang
Andrew Carroll
Anthony Khawaja
Babak Behsaz
American Journal of Human Genetics, 112 (2025), pp. 1562 - 1579
Preview abstract
Electronic health records, biobanks, and wearable biosensors enable the collection of multiple health modalities from many individuals. Access to multimodal health data provides a unique opportunity for genetic studies of complex traits because different modalities relevant to a single physiological system (e.g., circulatory system) encode complementary and overlapping information. We propose a multimodal deep learning method, multimodal representation learning for genetic discovery on low-dimensional embeddings (M-REGLE), for discovering genetic associations from a joint representation of complementary electrophysiological waveform modalities. M-REGLE jointly learns a lower representation (i.e., latent factors) of multimodal physiological waveforms using a convolutional variational autoencoder, performs genome-wide association studies (GWASs) on each latent factor, then combines the results to study the genetics of the underlying system. To validate the advantages of M-REGLE and multimodal learning, we apply it to common cardiovascular modalities (photoplethysmogram [PPG] and electrocardiogram [ECG]) and compare its results to unimodal learning methods in which representations are learned from each data modality separately but are statistically combined for downstream genetic comparison. M-REGLE identifies 19.3% more loci on the 12-lead ECG dataset, 13.0% more loci on the ECG lead I + PPG dataset, and its genetic risk score significantly outperforms the unimodal risk score at predicting cardiac phenotypes, such as atrial fibrillation (Afib), in multiple biobanks.
View details
GitChameleon 2.0: Evaluating AI Code Generation Against Python Library Version Incompatibilities
Diganta Misra
Nizar Islah
Brice Rauby
Zihan Wang
Justine Gehring
Antonio Orvieto
Muawiz Chaudhary
Eilif Muller
Irina Rish
Samira Ebrahimi Kahou
Massimo Caccia
2025
Preview abstract
The rapid evolution of software libraries poses a considerable hurdle for code generation, necessitating continuous adaptation to frequent version updates while preserving backward compatibility. While existing code evolution benchmarks provide valuable insights, they typically lack execution-based evaluation for generating code compliant with specific library versions. To address this, we introduce GitChameleon 2.0, a novel, meticulously curated dataset comprising 328 Python code completion problems, each conditioned on specific library versions and accompanied by executable unit tests. GitChameleon 2.0 rigorously evaluates the capacity of contemporary large language models (LLMs), LLM-powered agents, code assistants, and RAG systems to perform version-conditioned code generation that demonstrates functional accuracy through execution. Our extensive evaluations indicate that state-of-the-art systems encounter significant challenges with this task; enterprise models achieving baseline success rates in the 48-51% range, underscoring the intricacy of the problem. By offering an execution-based benchmark emphasizing the dynamic nature of code libraries, GitChameleon 2.0 enables a clearer understanding of this challenge and helps guide the development of more adaptable and dependable AI code generation methods.
View details
Preview abstract
Motivated by the growing demand for serving large language model inference requests, we study distributed load balancing for global serving systems with network latencies. We consider a fluid model in which continuous flows of requests arrive at different frontends and need to be routed to distant backends for processing whose processing rates are workload dependent. Network latencies can lead to long travel times for requests and delayed feedback from backends. The objective is to minimize the average latency of requests, composed of the network latency and the serving latency at the backends.
We introduce Distributed Gradient Descent Load Balancing (DGD-LB), a probabilistic routing algorithm in which each frontend adjusts the routing probabilities dynamically using gradient descent. Our algorithm is distributed: there is no coordination between frontends, except by observing the delayed impact other frontends have on shared backends. The algorithm uses an approximate gradient that measures the marginal impact of an additional request evaluated at a delayed system state. Equilibrium points of our algorithm minimize the centralized optimal average latencies, and we provide a novel local stability analysis showing that our algorithm converges to an optimal solution when started sufficiently close to that point. Moreover, we present sufficient conditions on the step-size of gradient descent that guarantee convergence in the presence of network latencies. Numerical experiments show that our algorithm is globally stable and optimal, confirm our stability conditions are nearly tight, and demonstrate that DGD-LB can lead to substantial gains relative to other load balancers studied in the literature when network latencies are large.
View details
I know what I don't know: improving model cascades through confidence tuning
Stephan Rabanser
Nathalie Rauschmayr
Petra Poklukar
Congchao Wang
2025
Preview abstract
Large-scale machine learning models deliver strong performance across a wide range of tasks but come with significant computational and resource constraints. To mitigate these challenges, local smaller models are often deployed alongside larger models, relying on routing and deferral mechanisms to offload complex tasks. However, existing approaches inadequately balance the capabilities of these models, often resulting in unnecessary deferrals or sub-optimal resource usage. In this work we introduce a novel loss function called Gatekeeper for calibrating smaller models in cascade setups. Our approach fine-tunes the smaller model to confidently handle tasks it can perform correctly while deferring complex tasks to the larger model. Moreover, it incorporates a mechanism for managing the trade-off between model performance and deferral accuracy, and is broadly applicable across various tasks and domains without any architectural changes. We evaluated our method on encoder-only, decoder-only, and encoder-decoder architectures. Experiments across image classification, language modeling, and vision-language tasks show that our approach substantially improves deferral performance.
View details
Preview abstract
Deep residual architectures, such as ResNet and the Transformer, have enabled models of unprecedented depth, yet a formal understanding of why depth is so effective remains an open question. A popular intuition, following Veit et al. (2016), is that these residual networks behave like ensembles of many shallower models. Our key finding is an explicit analytical formula that verifies this ensemble perspective, proving that increasing network depth is mathematically equivalent to expanding the size of this implicit ensemble. Furthermore, our expansion reveals a hierarchical ensemble structure in which the combinatorial growth of computation paths leads to an explosion in the output signal, explaining the historical necessity of normalization layers in training deep models. This insight offers a first principles explanation for the historical dependence on normalization layers and sheds new light on a family of successful normalization-free techniques like SkipInit and Fixup. However, while these previous approaches infer scaling factors through optimizer analysis or a heuristic analogy to Batch Normalization, our work offers the first explanation derived directly from the network's inherent functional structure. Specifically, our Residual Expansion Theorem reveals that scaling each residual module provides a principled solution to taming the combinatorial explosion inherent to these architectures. We further show that this scaling acts as a capacity controls that also implicitly regularizes the model's complexity.
View details
PLAN-TUNING: Post-Training Language Models to Learn Step-by-Step Planning for Complex Problem Solving
Mihir Parmar
Chitta Baral
Mingyang Ling
2025
Preview abstract
Recently, decomposing complex problems into simple subtasks--a crucial part of human-like natural planning--to solve the given problem has significantly boosted the performance of large language models (LLMs). However, leveraging such planning structures during post-training to boost the performance of smaller open-source LLMs remains underexplored. Motivated by this, we introduce Plan-Tuning, a unified post-training framework that (i) distills synthetic task decompositions (termed “planning trajectories”) from large-scale LLMs and (ii) fine-tunes smaller models via supervised and reinforcement-learning objectives designed to mimic these planning processes to improve complex reasoning. On GSM8k and the MATH benchmarks, plan-tuned models outperform strong baselines by an average ~7%. Furthermore, plan-tuned models show better generalization capabilities on out-of-domain datasets, with average ~10% and ~12% performance improvements on OlympiadBench and AIME 2024, respectively. Our detailed analysis demonstrates how planning trajectories improves complex reasoning capabilities, showing that Plan-Tuning is an effective strategy for improving task-specific performance of smaller LLMs.
View details
Syntactic and Semantic Gender Biases in the Language on Children’s Television: Evidence from a Corpus of 98 Shows from 1960 to 2018
Andrea Vial
Ruyuan Zuo
Shreya Havaldar
Morteza Dehghani
Andrei Cimpian
Psychological Science (2025)
Preview abstract
Biased media content shapes children’s social concepts and identities. We examined gender bias in a large corpus of scripts from 98 children’s television programs spanning 1960 to 2018 (6,600 episodes, ~2.7 million sentences, ~16 million words). We focused on agency and communion, the fundamental psychological dimensions underlying gender stereotypes. At the syntactic level, words referring to men/boys (vs. women/girls) appear more often in the agent (vs. patient) role. This syntactic bias remained stable between 1960 and 2018. At the semantic level, words referring to men/boys (vs. women/girls) co-occurred more often with words denoting agency. Words denoting communion showed both stereotypical and counterstereotypical associations. Some semantic gender biases have remained unchanged or weakened over time; others have grown. These findings suggest gender stereotypes are built into the core of children’s stories. Whether we are closer to gender equality in children’s media depends on where one looks.
View details
