publications

2026

1. AIES

   From Forensics to Ecosystems: Rethinking Watermarks for Generative AI Oversight

   Susser, Daniel, Thickstun, John, and Vidan, Gili

   In AI, Ethics, and Society 2026

   Abs arXiv

   The arrival of generative AI as a cheap, widely accessible commercial service, and the tidal wave of AI-generated synthetic content it has unleashed, have provoked deep epistemic and social anxieties and raised difficult governance questions that policymakers are struggling to address. One approach that has attracted both enthusiasm from regulators and skepticism from researchers is digital watermarking. Signals embedded in a synthetically-generated piece of content indicating that it was AI-generated—possibly even identifying the specific systems that generated it—appear to offer a path toward mitigating risks of genAI that avoids the downsides of more interventionist strategies. But critics warn that watermarks may prove technically brittle, epistemically ambiguous, and politically ineffectual tools. In this paper, we explore the challenges and opportunities of using digital watermarking for AI governance, paying special attention to the specific problem of watermarking AI-generated text. We argue that such critiques often treat the problem of identifying synthetic content as an isolated forensic question. Instead, we propose reconceptualizing digital watermarks as tools for understanding the impacts of synthetic content on media ecosystems, rather than reliably identifying individual pieces of synthetic content. Such an "ecosystems approach" more effectively utilizes the features of watermarks. And while this approach raises its own governance challenges, we argue that they are more tractable than the challenges of using watermarks for digital forensics.

2. ISMIR

   Assessing Factual Music Comprehension in Large Audio Language Models

   Lin, Daniel Chenyu, Freeman, Michael, and Thickstun, John

   In International Society for Music Information Retrieval 2026

   Abs arXiv

   Large audio language models (LALMs) leverage multimodal representations to generate open-ended answers to natural language queries about audio. In this paper, we (1) provide empirical evidence that assessment of LALMs using the popular MusicQA dataset fails to measure whether a model’s responses about music are factually correct, and (2) develop a new protocol for assessing the music comprehension capabilities of LALMs. Specifically, we propose an evaluation protocol that prompts a LALM for factually verifiable information, and parses its open-ended response into a structured format that can be objectively assessed using Precision, Recall, and F1 scores. Using this protocol, we define a benchmark consisting of six factual information retrieval tasks defined on three diverse datasets: MusicNet, the Free Music Archive, and OverClocked ReMix. We benchmark nine recent LALMs, including frontier models like Gemini and the latest open models like Music Flamingo, and release the suite of evaluation scripts at https://github.com/DCL2004/LALM-Eval to facilitate benchmarking of new LALMs.

3. ICML

   Scaling Beyond Masked Diffusion Language Models

   Sahoo, Subham Sekhar, Lemercier, Jean-Marie, Yang, Zhihan, Deschenaux, Justin, Liu, Jingyu, Thickstun, John, and Jukic, Ante

   In International Conference on Machine Learning 2026

   Abs arXiv Code Website

   Diffusion language models are a promising alternative to autoregressive models due to their potential for faster generation. Among discrete diffusion approaches, Masked diffusion currently dominates, largely driven by strong perplexity on language modeling benchmarks. In this work, we present the first scaling law study of uniform-state and interpolating discrete diffusion methods. We also show that Masked diffusion models can be made approximately 12% more FLOPs-efficient when trained with a simple cross-entropy objective. We find that perplexity is informative within a diffusion family but can be misleading across families, where models with worse likelihood scaling may be preferable due to faster and more practical sampling, as reflected by the speed-quality Pareto frontier. These results challenge the view that Masked diffusion is categorically the future of diffusion language modeling and that perplexity alone suffices for cross-algorithm comparison. Scaling all methods to 1.7B parameters, we show that uniform-state diffusion remains competitive on likelihood-based benchmarks and outperforms autoregressive and Masked diffusion models on GSM8K, despite worse validation perplexity.

4. ICML

   Esoteric Language Models: A Family of Any-Order Diffusion LLMs

   Sahoo, Subham Sekhar, Yang, Zhihan, Akhauri, Yash, Liu, Johnna, Singh, Deepansha, Cheng, Zhoujun, Liu, Zhengzhong, Xing, Eric P., Thickstun, John, and Vahdat, Arash

   In International Conference on Machine Learning 2026

   Abs arXiv Code Website

   Diffusion-based language models offer a compelling alternative to autoregressive (AR) models by enabling parallel and controllable generation. Within this family, Masked Diffusion Models (MDMs) currently perform best but still underperform AR models in perplexity and lack key inference-time efficiency features, most notably KV caching. We introduce Eso-LMs, a new family of models that fuses AR and MDM paradigms, smoothly interpolating between their perplexities while overcoming their respective limitations. Unlike prior work, which uses transformers with bidirectional attention as MDM denoisers, we exploit the connection between MDMs and Any-Order autoregressive models and adopt causal attention. This design lets us compute the exact likelihood of MDMs for the first time and, crucially, enables us to introduce KV caching for MDMs while preserving parallel generation for the first time, significantly improving inference efficiency.

5. ICLR-W

   Esoteric Language Models: Bridging Autoregressive and Masked Diffusion LLMs

   Sahoo, Subham Sekhar, Yang, Zhihan, Akhauri, Yash, Liu, Johnna, Singh, Deepansha, Cheng, Zhoujun, Liu, Zhengzhong, Xing, Eric, Thickstun, John, and Vahdat, Arash

   In ICLR Workshop on Multimodal Intelligence: Next Token Prediction and Beyond 2026

   Abs arXiv PDF Code Website

   Diffusion-based language models offer a compelling alternative to autoregressive (AR) models by enabling parallel and controllable generation. Within this family, Masked Diffusion Models (MDMs) currently perform best but still underperform AR models in perplexity and lack key inference-time efficiency features, most notably KV caching. We introduce Esoteric Language Models (Eso-LMs), a new family of models that fuses AR and MDM paradigms, smoothly interpolating between their perplexities while overcoming their respective limitations. Unlike prior work, which uses transformers with bidirectional attention as MDM denoisers, we exploit the connection between MDMs and Any-Order autoregressive models and adopt causal attention. This design lets us compute the exact likelihood of MDMs for the first time and, crucially, enables us to introduce exact KV caching for MDMs while preserving parallel generation over the full sequence length for the first time, significantly improving inference efficiency. Combined with an optimized sampling schedule, Eso-LMs achieves a new state of the art on the speed-quality Pareto frontier for unconditional generation. On longer contexts, it yields 14 − 65× faster inference than standard MDMs and 3 − 4× faster inference than prior semi-autoregressive approaches. We provide code, model checkpoints, and a talk recording on the project page: https://s-sahoo.com/Eso-LMs.

6. ICLR-W

   The Reliability Gap in Agentic Evidence Verification for Materials Science

   Gong, Albert, Kim, James J., Kabra, Anmol, Panigrahi, Aaditya, Wang, Jiashuo, Mulchandani, Arjun B., Freeman, Michael, Katmer, Fatmagul, Wakefield, Joshua Peters, Zhao, Linxi, Wan, Chao, Sarkar, Akanksha, Artzi, Yoav, Schoop, Leslie M., Thickstun, John, Weinberger, Kilian Q., Kim, Eun-Ah, Frazier, Peter I., and Sun, Jennifer J.

   In ICLR Workshops on Agents in the Wild (AIWILD) and Foundation Models for Science (FM4Science) 2026

   Abs

   The explosive growth of scientific literature has created an urgent need for autonomous agents capable of rigorous evidence verification. However, existing evaluation frameworks largely rely on sanitized oracle settings (e.g. pre-parsed PDFs) or controlled knowledge environments (e.g. curated corpus), which abstracts away the noise and complexity of real-world research. We develop a framework for evaluating agentic evidence verification under realistic conditions, grounding our study on a set of 200 open-source papers from the SuperCon database. We select materials science as our testbed because it demands exact numerical grounding and binary factual certainty—strict constraints that expose model reliability issues. We develop our evaluation criteria on accuracy and reliability with material science experts, focusing on two core workflows: Multi-Property Extraction from raw, multimodal PDFs, and Open-World Precedent Search on the live web, explicitly testing the ability to verify "negative results". Systematic study of general-purpose agents reveals a sobering reality: models often identify relevant properties or materials but fail to reliably extract precise values or ensure that answer correctness is grounded with valid attribution—highlighting a fundamental reliability gap in current systems.

7. KDD

   Benchmark Datasets for Lead-Lag Forecasting on Social Platforms

   Kazemian, Kimia, Liu, Zhenzhen, Yang, Yangfanyu, Luo, Katie Z., Gu, Shuhan, Du, Audrey, Yang, Xinyu, Jansons, Jack, Weinberger, Kilian Q., Thickstun, John, Yin, Yian, and Dean, Sarah

   In Knowledge Discovery in Databases - Datasets and Benchmarks Track 2026

   Abs arXiv Website

   Social and collaborative platforms emit multivariate time-series traces in which early interactions – such as views, likes, or downloads – are followed, sometimes months or years later, by higher impact like citations, sales, or reviews. We formalize this setting as Lead-Lag Forecasting (LLF): given an early usage channel (the lead), predict a correlated but temporally shifted outcome channel (the lag). Despite the ubiquity of such patterns, LLF has not been treated as a unified forecasting problem within the time-series community, largely due to the absence of standardised datasets. To anchor research in LLF, here we present two high-volume benchmark datasets: arXiv (accesses to citations of 2.3M papers) and GitHub (pushes/stars to forks of 3M repositories). Our datasets provide ideal testbeds for lead-lag forecasting, by capturing long-horizon dynamics across years, spanning the full spectrum of outcomes, and avoiding survivorship bias in sampling. We documented all technical details of data curation and cleaning, verified the presence of lead-lag dynamics through statistical and classification tests, and benchmarked parametric and non-parametric baselines for regression.

2025

1. NeurIPS-W

   Robust Neural Audio Fingerprinting using Music Foundation Models

   Singh, Shubhr, Bhat, Kiran, Resnick, Benjamin, Riley, Xavier, Thickstun, John, and Brouwer, Walter De

   In NeurIPS Workshop on AI for Music 2025

   Abs arXiv

   The proliferation of distorted, compressed, and manipulated music on modern media platforms like TikTok motivates the development of more robust audio fingerprinting techniques to identify the sources of musical recordings. In this paper, we develop and evaluate new neural audio fingerprinting techniques with the aim of improving their robustness. We make two contributions to neural fingerprinting methodology: (1) we use a pretrained music foundation model as the backbone of the neural architecture and (2) we expand the use of data augmentation to train fingerprinting models under a wide variety of audio manipulations, including time stretching, pitch modulation, compression, and filtering. We systematically evaluate our methods in comparison to two state-of-the-art neural fingerprinting models: NAFP and GraFPrint. Results show that fingerprints extracted with music foundation models (e.g., MuQ, MERT) consistently outperform models trained from scratch or pretrained on non-musical audio. Segment-level evaluation further reveals their capability to accurately localize fingerprint matches, an important practical feature for catalog management.

2. Neurips

   Linearly Constrained Diffusion Implicit Models

   Jayaram, Vivek, Kemelmacher-Shlizerman, Ira, Seitz, Steven M, and Thickstun, John

   In Advances in Neural Information Processing Systems 2025

   Abs arXiv Code Website

   We introduce Linearly Constrained Diffusion Implicit Models (CDIM), a fast and accurate approach to solving noisy linear inverse problems using diffusion models. Traditional diffusion-based inverse methods rely on numerous projection steps to enforce measurement consistency in addition to unconditional denoising steps. CDIM achieves a 10-50x reduction in projection steps by dynamically adjusting the number and size of projection steps to align a residual measurement energy with its theoretical distribution under the forward diffusion process. This adaptive alignment preserves measurement consistency while substantially accelerating constrained inference. For noise-free linear inverse problems, CDIM exactly satisfies the measurement constraints with few projection steps, even when existing methods fail. We demonstrate CDIM’s effectiveness across a range of applications, including super-resolution, denoising, inpainting, deblurring, and 3D point cloud reprojection.

3. ISMIR

   Aligning Text-to-Music Evaluation with Human Preferences

   Huang, Yichen, Novack, Zachary, Saito, Koichi, Shi, Jiatong, Watanabe, Shinji, Mitsufuji, Yuki, Thickstun, John, and Donahue, Chris

   In International Society for Music Information Retrieval 2025

   Abs arXiv Code

   Despite significant recent advances in generative acoustic text-to-music (TTM) modeling, robust evaluation of these models lags behind, relying in particular on the popular Fréchet Audio Distance (FAD). In this work, we rigorously study the design space of reference-based divergence metrics for evaluating TTM models through (1) designing four synthetic meta-evaluations to measure sensitivity to particular musical desiderata, and (2) collecting and evaluating on MusicPrefs, the first open-source dataset of human preferences for TTM systems. We find that not only is the standard FAD setup inconsistent on both synthetic and human preference data, but that nearly all existing metrics fail to effectively capture desiderata, and are only weakly correlated with human perception. We propose a new metric, the MAUVE Audio Divergence (MAD), computed on representations from a self-supervised audio embedding model. We find that this metric effectively captures diverse musical desiderata (average rank correlation 0.84 for MAD vs. 0.49 for FAD and also correlates more strongly with MusicPrefs (0.62 vs. 0.14).

2024

1. ISMIR-LBD

   Hookpad Aria: A Copilot for Songwriters

   Donahue, Chris, Wu, Shih-Lun, Kim, Yewon, Carlton, Dave, Miyakawa, Ryan, and Thickstun, John

   In International Society for Music Information Retrieval Late Breaking Demos 2024

   Abs arXiv Website

   Recent progress in generative models for music has highlighted the need for interactive, controllable systems that service the goals of musicians. In this paper, we introduce a system that integrates a compositional assistant and live improvisational partner directly into the modern music producer’s toolkit–the digital audio workstation. Our system design is guided by 1) integration with modern music production software, 2) non-linear songwriting workflows, and 3) enabling generative AI to support live improvisation tasks. We find that anticipatory transformer models are well suited for these goals, and present a method for adapting an anticipatory model for live improvisation. We call on future work to further explore human-AI co-performance by designing systems to be accessible and integrated into the workflows of domain experts.

2. GenAICHI

   Designing Live Human-AI Collaboration for Musical Improvisation

   Becker, Nic, Louie, Ryan, Thickstun, John, and Liang, Percy

   In CHI Workshop on Generative AI and HCI 2024

   Abs PDF

   Recent progress in generative models for music has highlighted the need for interactive, controllable systems that service the goals of musicians. In this paper, we introduce a system that integrates a compositional assistant and live improvisational partner directly into the modern music producer’s toolkit–the digital audio workstation. Our system design is guided by 1) integration with modern music production software, 2) non-linear songwriting workflows, and 3) enabling generative AI to support live improvisation tasks. We find that anticipatory transformer models are well suited for these goals, and present a method for adapting an anticipatory model for live improvisation. We call on future work to further explore human-AI co-performance by designing systems to be accessible and integrated into the workflows of domain experts.

3. TMLR

   Robust distortion-free watermarks for language models

   Kuditipudi, Rohith, Thickstun, John, Hashimoto, Tatsunori, and Liang, Percy

   Transactions on Machine Learning Research 2024

   Abs arXiv Code Website Talk

   We propose a methodology for planting watermarks in text from an autoregressive language model that are robust to perturbations without changing the distribution over text up to a certain maximum generation budget. We generate watermarked text by mapping a sequence of random numbers – which we compute using a randomized watermark key – to a sample from the language model. To detect watermarked text, any party who knows the key can align the text to the random number sequence. We instantiate our watermark methodology with two sampling schemes: inverse transform sampling and exponential minimum sampling. We apply these watermarks to three language models – OPT-1.3B, LLaMA-7B and Alpaca-7B – to experimentally validate their statistical power and robustness to various paraphrasing attacks. Notably, for both the OPT-1.3B and LLaMA-7B models, we find we can reliably detect watermarked text (p≤0.01) from 35 tokens even after corrupting between 40-50% of the tokens via random edits (i.e., substitutions, insertions or deletions). For the Alpaca-7B model, we conduct a case study on the feasibility of watermarking responses to typical user instructions. Due to the lower entropy of the responses, detection is more difficult: around 25% of the responses – whose median length is around 100 tokens – are detectable with p≤0.01, and the watermark is also less robust to certain automated paraphrasing attacks we implement.

4. TMLR

   Anticipatory music transformer

   Thickstun, John, Hall, David, Donahue, Chris, and Liang, Percy

   Transactions on Machine Learning Research 2024

   Abs arXiv Code Website Media Talk

   We introduce anticipation: a method for constructing a controllable generative model of a temporal point process (the event process) conditioned asynchronously on realizations of a second, correlated process (the control process). We achieve this by interleaving sequences of events and controls, such that controls appear following stopping times in the event sequence. This work is motivated by problems arising in the control of symbolic music generation. We focus on infilling control tasks, whereby the controls are a subset of the events themselves, and conditional generation completes a sequence of events given the fixed control events. We train anticipatory infilling models using the large and diverse Lakh MIDI music dataset. These models match the performance of autoregressive models for prompted music generation, with the additional capability to perform infilling control tasks, including accompaniment. Human evaluators report that an anticipatory model produces accompaniments with similar musicality to even music composed by humans over a 20-second clip.

2023

1. JMLR

   MAUVE Scores for Generative Models: Theory and Practice

   Pillutla, Krishna, Liu, Lang, Thickstun, John, Welleck, Sean, Swayamdipta, Swabha, Zellers, Rowan, Oh, Sewoong, Choi, Yejin, and Harchaoui, Zaid

   Journal of Machine Learning Research 2023

   Abs arXiv Code

   Generative AI has matured to a point where large-scale models can generate text that seems indistinguishable from human-written text and remarkably photorealistic images. Automatically measuring how close the distribution of generated data is to the target real data distribution is a key step in diagnosing existing models and developing better models. We present MAUVE, a family of comparison measures between pairs of distributions such as those encountered in the generative modeling of text or images. These scores are statistical summaries of divergence frontiers capturing two types of errors in generative modeling. We explore four approaches to statistically estimate these scores: vector quantization, non-parametric estimation, classifier-based estimation, and parametric Gaussian approximations. We provide statistical bounds for the vector quantization approach. Empirically, we find that the proposed scores paired with a range of f-divergences and statistical estimation methods can quantify the gaps between the distributions of human-written text and those of modern neural language models by correlating with human judgments and identifying known properties of the generated texts. We conclude the paper by demonstrating its applications to other AI domains and discussing practical recommendations.

2. ACL Outstanding Paper

   Backpack language models

   Hewitt, John, Thickstun, John, Manning, Christopher D., and Liang, Percy

   In Proceedings of the Association for Computational Linguistics 2023

   Abs arXiv Code Website

   We present Backpacks: a new neural architecture that marries strong modeling performance with an interface for interpretability and control. Backpacks learn multiple non-contextual sense vectors for each word in a vocabulary, and represent a word in a sequence as a context-dependent, non-negative linear combination of sense vectors in this sequence. We find that, after training, sense vectors specialize, each encoding a different aspect of a word. We can interpret a sense vector by inspecting its (non-contextual, linear) projection onto the output space, and intervene on these interpretable hooks to change the model’s behavior in predictable ways. We train a 170M-parameter Backpack language model on OpenWebText, matching the loss of a GPT-2 small (124M parameter) Transformer. On lexical similarity evaluations, we find that Backpack sense vectors outperform even a 6B-parameter Transformer LM’s word embeddings. Finally, we present simple algorithms that intervene on sense vectors to perform controllable text generation and debiasing. For example, we can edit the sense vocabulary to tend more towards a topic, or localize a source of gender bias to a sense vector and globally suppress that sense.

3. TMLR

   Evaluating Human-Language Model Interaction

   Lee, Mina, Srivastava, Megha, Hardy, Amelia, Thickstun, John, Durmus, Esin, Paranjape, Ashwin, Gerard-Ursin, Ines, Li, Xiang Lisa, Ladhak, Faisal, Rong, Frieda, Wang, Rose E., Kwon, Minae, Park, Joon Sung, Cao, Hancheng, Lee, Tony, Bommasani, Rishi, Bernstein, Michael, and Liang, Percy

   Transactions on Machine Learning Research 2023

   Abs arXiv Code

   Many real-world applications of language models (LMs), such as code autocomplete and writing assistance, involve human-LM interaction, but the main LM benchmarks are non-interactive, where a system produces output without human intervention. To evaluate human-LM interaction, we develop a framework, Human-AI Language-based Interaction Evaluation (H-LINE), that expands non-interactive evaluation along three dimensions, capturing (i) the interactive process, not only the final output; (ii) the first-person subjective experience, not just a third-party assessment; and (iii) notions of preference beyond quality. We then design five tasks ranging from goal-oriented to open-ended to capture different forms of interaction. On four state-of-the-art LMs (three variants of OpenAI’s GPT-3 and AI21’s J1-Jumbo), we find that non-interactive performance does not always result in better human-LM interaction and that first-person and third-party metrics can diverge, suggesting the importance of examining the nuances of human-LM interaction.

2022

1. ISMIR

   Melody transcription via generative pre-training

   Donahue, Chris, Thickstun, John, and Liang, Percy

   In International Society for Music Information Retrieval 2022

   Abs arXiv Code Website

   Despite the central role that melody plays in music perception, it remains an open challenge in music information retrieval to reliably detect the notes of the melody present in an arbitrary music recording. A key challenge in melody transcription is building methods which can handle broad audio containing any number of instrument ensembles and musical styles - existing strategies work well for some melody instruments or styles but not all. To confront this challenge, we leverage representations from Jukebox (Dhariwal et al. 2020), a generative model of broad music audio, thereby improving performance on melody transcription by 20% relative to conventional spectrogram features. Another obstacle in melody transcription is a lack of training data - we derive a new dataset containing 50 hours of melody transcriptions from crowdsourced annotations of broad music. The combination of generative pre-training and a new dataset for this task results in 77% stronger performance on melody transcription relative to the strongest available baseline. By pairing our new melody transcription approach with solutions for beat detection, key estimation, and chord recognition, we build Sheet Sage, a system capable of transcribing human-readable lead sheets directly from music audio.

2. bioRxiv

   Reconstruction of visual images from mouse retinal ganglion cell spiking activity using convolutional neural networks

   Benster, Tyler, Babino, Darwin, Thickstun, John, Hunt, Matthew, Liu, Xiyang, Harchaoui, Zaid, Oh, Sewoong, and Gelder, Russell N Van

   2022

   Abs PDF Code

   All visual information in mammals is encoded in the aggregate pattern of retinal ganglion cell (RGC) firing. How this information is decoded to yield percepts remains incompletely understood. We have trained convolutional neural networks with multielectrode array-recorded murine RGC responses to projected images. The trained model accurately reconstructed novel facial images solely from RGC firing data. In this model, subpopulations of cells with faster firing rates are largely sufficient for accurate reconstruction, and ON- and OFF-cells contribute complementary and overlapping information to image reconstruction. Information content for reconstruction correlates with overall firing rate, and locality of information contributing to reconstruction varies substantially across the image and retina. This model demonstrates that artificial neural networks are capable of learning multicellular sensory neural encoding, and provides a viable model for understanding visual information encoding.

3. Neurips Oral Presentation

   Diffusion-LM improves controllable text generation

   Li, Xiang Lisa, Thickstun, John, Gulrajani, Ishaan, Liang, Percy, and Hashimoto, Tatsunori B.

   In Advances in Neural Information Processing Systems 2022

   Abs arXiv Code Slides

   Controlling the behavior of language models (LMs) without re-training is a major open problem in natural language generation. While recent works have demonstrated successes on controlling simple sentence attributes (e.g., sentiment), there has been little progress on complex, fine-grained controls (e.g., syntactic structure). To address this challenge, we develop a new non-autoregressive language model based on continuous diffusions that we call Diffusion-LM. Building upon the recent successes of diffusion models in continuous domains, Diffusion-LM iteratively denoises a sequence of Gaussian vectors into word vectors, yielding a sequence of intermediate latent variables. The continuous, hierarchical nature of these intermediate variables enables a simple gradient-based algorithm to perform complex, controllable generation tasks. We demonstrate successful control of Diffusion-LM for six challenging fine-grained control tasks, significantly outperforming prior work.

2021

1. Dissertation

   Leveraging generative models for music and signal processing

   Thickstun, John

   University of Washington 2021

   PDF
2. Neurips Outstanding Paper

   MAUVE: measuring the gap between neural text and human text using divergence frontiers

   Pillutla, Krishna, Swayamdipta, Swabha, Zellers, Rowan, Thickstun, John, Welleck, Sean, Choi, Yejin, and Harchaoui, Zaid

   In Advances in Neural Information Processing Systems 2021

   Abs arXiv Code Slides

   As major progress is made in open-ended text generation, measuring how close machine-generated text is to human language remains a critical open problem. We introduce MAUVE, a comparison measure for open-ended text generation, which directly compares the learnt distribution from a text generation model to the distribution of human-written text using divergence frontiers. MAUVE scales up to modern text generation models by computing information divergences in a quantized embedding space. Through an extensive empirical study on three open-ended generation tasks, we find that MAUVE identifies known properties of generated text, scales naturally with model size, and correlates with human judgments, with fewer restrictions than existing distributional evaluation metrics.

3. ICML

   Parallel and flexible sampling from autoregressive models via Langevin dynamics

   Jayaram, Vivek, and Thickstun, John

   In International Conference on Machine Learning 2021

   Abs arXiv Code Poster Website

   This paper introduces an alternative approach to sampling from autoregressive models. Autoregressive models are typically sampled sequentially, according to the transition dynamics defined by the model. Instead, we propose a sampling procedure that initializes a sequence with white noise and follows a Markov chain defined by Langevin dynamics on the global log-likelihood of the sequence. This approach parallelizes the sampling process and generalizes to conditional sampling. Using an autoregressive model as a Bayesian prior, we can steer the output of a generative model using a conditional likelihood or constraints. We apply these techniques to autoregressive models in the visual and audio domains, with competitive results for audio source separation, super-resolution, and inpainting.

4. L4DC

   Faster policy learning with continuous-time gradients

   Ainsworth, Samuel K., Lowrey, Kendall, Thickstun, John, Harchaoui, Zaid, and Srinivasa, Siddhartha S.

   In Learning for Dynamics and Control 2021

   Abs arXiv Code

   We study the estimation of policy gradients for continuous-time systems with known dynamics. By reframing policy learning in continuous-time, we show that it is possible construct a more efficient and accurate gradient estimator. The standard back-propagation through time estimator (BPTT) computes exact gradients for a crude discretization of the continuous-time system. In contrast, we approximate continuous-time gradients in the original system. With the explicit goal of estimating continuous-time gradients, we are able to discretize adaptively and construct a more efficient policy gradient estimator which we call the Continuous-Time Policy Gradient (CTPG). We show that replacing BPTT policy gradients with more efficient CTPG estimates results in faster and more robust learning in a variety of control tasks and simulators.

2020

1. arXiv

   Rethinking evaluation methodology for audio-to-score alignment

   Thickstun, John, Brennan, Jennifer, and Verma, Harsh

   arXiv preprint arXiv:2009.14374 2020

   Abs arXiv Code

   This paper offers a precise, formal definition of an audio-to-score alignment. While the concept of an alignment is intuitively grasped, this precision affords us new insight into the evaluation of audio-to-score alignment algorithms. Motivated by these insights, we introduce new evaluation metrics for audio-to-score alignment. Using an alignment evaluation dataset derived from pairs of KernScores and MAESTRO performances, we study the behavior of our new metrics and the standard metrics on several classical alignment algorithms.

2. EMNLP

   An information bottleneck approach for controlling conciseness in rationale extraction

   Paranjape, Bhargavi, Joshi, Mandar, Thickstun, John, Hajishirzi, Hannaneh, and Zettlemoyer, Luke

   In Conference on Empirical Methods in Natural Language Processing 2020

   Abs arXiv Code Slides

   Decisions of complex language understanding models can be rationalized by limiting their inputs to a relevant subsequence of the original text. A rationale should be as concise as possible without significantly degrading task performance, but this balance can be difficult to achieve in practice. In this paper, we show that it is possible to better manage this trade-off by optimizing a bound on the Information Bottleneck (IB) objective. Our fully unsupervised approach jointly learns an explainer that predicts sparse binary masks over sentences, and an end-task predictor that considers only the extracted rationale. Using IB, we derive a learning objective that allows direct control of mask sparsity levels through a tunable sparse prior. Experiments on ERASER benchmark tasks demonstrate significant gains over norm-minimization techniques for both task performance and agreement with human rationales. Furthermore, we find that in the semi-supervised setting, a modest amount of gold rationales (25% of training examples) closes the gap with a model that uses the full input.

3. ICML

   Source separation with deep generative priors

   Jayaram, Vivek, and Thickstun, John

   In International Conference on Machine Learning 2020

   Abs arXiv Code Poster

   Despite substantial progress in signal source separation, results for richly structured data continue to contain perceptible artifacts. In contrast, recent deep generative models can produce authentic samples in a variety of domains that are indistinguishable from samples of the data distribution. This paper introduces a Bayesian approach to source separation that uses generative models as priors over the components of a mixture of sources, and noise-annealed Langevin dynamics to sample from the posterior distribution of sources given a mixture. This decouples the source separation problem from generative modeling, enabling us to directly use cutting-edge generative models as priors. The method achieves state-of-the-art performance for MNIST digit separation. We introduce new methodology for evaluating separation quality on richer datasets, providing quantitative evaluation of separation results on CIFAR-10. We also provide qualitative results on LSUN.

2019

1. ISMIR

   Convolutional composer classification

   Verma, Harsh, and Thickstun, John

   In International Society for Music Information Retrieval 2019

   Abs arXiv Code Poster

   This paper investigates end-to-end learnable models for attributing composers to musical scores. We introduce several pooled, convolutional architectures for this task and draw connections between our approach and classical learning approaches based on global and n-gram features. We evaluate models on a corpus of 2,500 scores from the KernScores collection, authored by a variety of composers spanning the Renaissance era to the early 20th century. This corpus has substantial overlap with the corpora used in several previous, smaller studies; we compare our results on subsets of the corpus to these previous works.

2. ISMIR

   Coupled recurrent models for polyphonic music composition

   Thickstun, John, Harchaoui, Zaid, Foster, Dean P, and Kakade, Sham M

   In International Society for Music Information Retrieval 2019

   Abs arXiv Code Poster

   This paper introduces a novel recurrent model for music composition that is tailored to the structure of polyphonic music. We propose an efficient new conditional probabilistic factorization of musical scores, viewing a score as a collection of concurrent, coupled sequences: i.e. voices. To model the conditional distributions, we borrow ideas from both convolutional and recurrent neural models; we argue that these ideas are natural for capturing music’s pitch invariances, temporal structure, and polyphony. We train models for single-voice and multi-voice composition on 2,300 scores from the KernScores dataset.

2018

1. ICASSP Oral Presentation

   Invariances and data augmentation for supervised music transcription

   Thickstun, John, Harchaoui, Zaid, Foster, Dean P, and Kakade, Sham M

   In International Conference on Acoustics, Speech and Signal Processing 2018

   Abs arXiv Code

   This paper explores a variety of models for frame-based music transcription, with an emphasis on the methods needed to reach state-of-the-art on human recordings. The translation-invariant network discussed in this paper, which combines a traditional filterbank with a convolutional neural network, was the top-performing model in the 2017 MIREX Multiple Fundamental Frequency Estimation evaluation. This class of models shares parameters in the log-frequency domain, which exploits the frequency invariance of music to reduce the number of model parameters and avoid overfitting to the training data. All models in this paper were trained with supervision by labeled data from the MusicNet dataset, augmented by random label-preserving pitch-shift transformations.

2017

1. MIREX

   Frequency domain convolutions for multiple F0 estimation

   Thickstun, John, Harchaoui, Zaid, Foster, Dean P, and Kakade, Sham M

   2017

   Abs PDF

   This document describes the THK1 submission in the 2017 MIREX Multi-F0 competition. The model is a convolutional neural network, trained using the MusicNet labels. Its input is a bank of logarithmically-spaced frequency filters. These filters exhibit translation invariance along the log-frequency axis, which is captured in this model by one-dimensional convolutions along the frequency axis. The model fully connects across the time axis to capture temporal dependencies. The training data is augmented by pitch-shifting the original data by up to 5 semitones and applying small (up to 1/10 semitone) continuous pitch jitter to the input.

2. ICLR

   Learning features of music from scratch

   Thickstun, John, Harchaoui, Zaid, and Kakade, Sham M

   In International Conference on Learning Representations 2017

   Abs arXiv Code Poster Website

   This paper introduces a new large-scale music dataset, MusicNet, to serve as a source of supervision and evaluation of machine learning methods for music research. MusicNet consists of hundreds of freely-licensed classical music recordings by 10 composers, written for 11 instruments, together with instrument/note annotations resulting in over 1 million temporal labels on 34 hours of chamber music performances under various studio and microphone conditions. The paper defines a multi-label classification task to predict notes in musical recordings, along with an evaluation protocol, and benchmarks several machine learning architectures for this task: i) learning from spectrogram features; ii) end to-end learning with a neural net; iii) end-to-end learning with a convolutional neural net. These experiments show that end-to-end models trained for note prediction learn frequency selective filters as a low-level representation of audio.