NeurIPS 2023 Papers

While recent large-scale neural codec language models have shown significant improvement in zero-shot TTS by training on thousands of hours of data, they suffer from drawbacks such as a lack of robustness, slow sampling speed similar to previous autoregressive TTS methods, and reliance on pre-trained neural codec representations. Our work proposes P-Flow, a fast and data-efficient zero-shot TTS model that uses speech prompts for speaker adaptation. P-Flow comprises a speech-prompted text encoder for speaker adaptation and a flow matching generative decoder for high-quality and fast speech synthesis. Our speech-prompted text encoder uses speech prompts and text input to generate speaker-conditional text representation. The flow matching generative decoder uses the speaker-conditional output to synthesize high-quality personalized speech significantly faster than in real-time. Unlike the neural codec language models, we specifically train P-Flow on LibriTTS dataset using a continuous mel-representation. Through our training method using continuous speech prompts, P-Flow matches the speaker similarity performance of the large-scale zero-shot TTS models with two orders of magnitude less training data and has more than 20$\times$ faster sampling speed. Our results show that P-Flow has better pronunciation and is preferred in human likeness and speaker similarity to its recent state-of-the-art counterparts, thus defining P-Flow as an attractive and desirable alternative. We provide audio samples on our demo page: [https://research.nvidia.com/labs/adlr/projects/pflow](https://research.nvidia.com/labs/adlr/projects/pflow)

Solving complicated AI tasks with different domains and modalities is a key step toward artificial general intelligence. While there are numerous AI models available for various domains and modalities, they cannot handle complicated AI tasks autonomously. Considering large language models (LLMs) have exhibited exceptional abilities in language understanding, generation, interaction, and reasoning, we advocate that LLMs could act as a controller to manage existing AI models to solve complicated AI tasks, with language serving as a generic interface to empower this. Based on this philosophy, we present HuggingGPT, an LLM-powered agent that leverages LLMs (e.g., ChatGPT) to connect various AI models in machine learning communities (e.g., Hugging Face) to solve AI tasks. Specifically, we use ChatGPT to conduct task planning when receiving a user request, select models according to their function descriptions available in Hugging Face, execute each subtask with the selected AI model, and summarize the response according to the execution results. By leveraging the strong language capability of ChatGPT and abundant AI models in Hugging Face, HuggingGPT can tackle a wide range of sophisticated AI tasks spanning different modalities and domains and achieve impressive results in language, vision, speech, and other challenging tasks, which paves a new way towards the realization of artificial general intelligence.

Deep neural networks are notorious for defying theoretical treatment. However, when the number of parameters in each layer tends to infinity, the network function is a Gaussian process (GP) and quantitatively predictive description is possible. Gaussian approximation allows one to formulate criteria for selecting hyperparameters, such as variances of weights and biases, as well as the learning rate. These criteria rely on the notion of criticality defined for deep neural networks. In this work we describe a new practical way to diagnose criticality. We introduce *partial Jacobians* of a network, defined as derivatives of preactivations in layer $l$ with respect to preactivations in layer $l_0\leq l$. We derive recurrence relations for the norms of partial Jacobians and utilize these relations to analyze criticality of deep fully connected neural networks with LayerNorm and/or residual connections. We derive and implement a simple and cheap numerical test that allows one to select optimal initialization for a broad class of deep neural networks; containing fully connected, convolutional and normalization layers. Using these tools we show quantitatively that proper stacking of the LayerNorm (applied to preactivations) and residual connections leads to an architecture that is critical for any initialization. Finally, we apply our methods to analyze ResNet and MLP-Mixer architectures; demonstrating the everywhere-critical regime.

Knowledge distillation is commonly used to compress an ensemble of models into a single model. In this work we study the problem of progressive ensemble distillation: Given a large, pretrained teacher model , we seek to decompose the model into an ensemble of smaller, low-inference cost student models . The resulting ensemble allows for flexibly tuning accuracy vs. inference cost, which can be useful for a multitude of applications in efficient inference. Our method, B-DISTIL, uses a boosting procedure that allows function composition based aggregation rules to construct expressive ensembles with similar performance as using much smaller student models. We demonstrate the effectiveness of B-DISTIL by decomposing pretrained models across a variety of image, speech, and sensor datasets. Our method comes with strong theoretical guarantees in terms of convergence as well as generalization.

In this paper, we introduce self-distillation and online clustering for self-supervised speech representation learning (DinoSR) which combines masked language modeling, self-distillation, and online clustering. We show that these concepts complement each other and result in a strong representation learning model for speech. DinoSR first extracts contextualized embeddings from the input audio with a teacher network, then runs an online clustering system on the embeddings to yield a machine-discovered phone inventory, and finally uses the discretized tokens to guide a student network. We show that DinoSR surpasses previous state-of-the-art performance in several downstream tasks, and provide a detailed analysis of the model and the learned discrete units.

Innovation is a major driver of economic and social development, and information about many kinds of innovation is embedded in semi-structured data from patents and patent applications. Though the impact and novelty of innovations expressed in patent data are difficult to measure through traditional means, machine learning offers a promising set of techniques for evaluating novelty, summarizing contributions, and embedding semantics. In this paper, we introduce the Harvard USPTO Patent Dataset (HUPD), a large-scale, well-structured, and multi-purpose corpus of English-language patent applications filed to the United States Patent and Trademark Office (USPTO) between 2004 and 2018. With more than 4.5 million patent documents, HUPD is two to three times larger than comparable corpora. Unlike other NLP patent datasets, HUPD contains the inventor-submitted versions of patent applications, not the final versions of granted patents, allowing us to study patentability at the time of filing using NLP methods for the first time. It is also novel in its inclusion of rich structured data alongside the text of patent filings: By providing each application’s metadata along with all of its text fields, HUPD enables researchers to perform new sets of NLP tasks that leverage variation in structured covariates. As a case study on the types of research HUPD makes possible, we introduce a new task to the NLP community -- patent acceptance prediction. We additionally show the structured metadata provided in HUPD allows us to conduct explicit studies of concept shifts for this task. We find that performance on patent acceptance prediction decays when models trained in one context are evaluated on different innovation categories and over time. Finally, we demonstrate how HUPD can be used for three additional tasks: Multi-class classification of patent subject areas, language modeling, and abstractive summarization. Put together, our publicly-available dataset aims to advance research extending language and classification models to diverse and dynamic real-world data distributions.

A popular heuristic method for improving clustering results is to apply dimensionality reduction before running clustering algorithms. It has been observed that spectral-based dimensionality reduction tools, such as PCA or SVD, improve the performance of clustering algorithms in many applications. This phenomenon indicates that spectral method not only serves as a dimensionality reduction tool, but also contributes to the clustering procedure in some sense. It is an interesting question to understand the behavior of spectral steps in clustering problems. As an initial step in this direction, this paper studies the power of vanilla-SVD algorithm in the stochastic block model (SBM). We show that, in the symmetric setting, vanilla-SVD algorithm recovers all clusters correctly. This result answers an open question posed by Van Vu (Combinatorics Probability and Computing, 2018) in the symmetric setting.

Vision and text have been fully explored in contemporary video-text foundational models, while other modalities such as audio and subtitles in videos have not received sufficient attention. In this paper, we resort to establish connections between multi-modality video tracks, including Vision, Audio, and Subtitle, and Text by exploring an automatically generated large-scale omni-modality video caption dataset called VAST-27M. Specifically, we first collect 27 million open-domain video clips and separately train a vision and an audio captioner to generate vision and audio captions. Then, we employ an off-the-shelf Large Language Model (LLM) to integrate the generated captions, together with subtitles and instructional prompts into omni-modality captions. Based on the proposed VAST-27M dataset, we train an omni-modality video-text foundational model named VAST, which can perceive and process vision, audio, and subtitle modalities from video, and better support various tasks including vision-text, audio-text, and multi-modal video-text tasks (retrieval, captioning and QA). Extensive experiments have been conducted to demonstrate the effectiveness of our proposed VAST-27M corpus and VAST foundation model. VAST achieves 22 new state-of-the-art results on various cross-modality benchmarks.

Recently, the Deep Learning community has become interested in evolutionary optimization (EO) as a means to address hard optimization problems, e.g. meta-learning through long inner loop unrolls or optimizing non-differentiable operators. One core reason for this trend has been the recent innovation in hardware acceleration and compatible software -- making distributed population evaluations much easier than before. Unlike for gradient descent-based methods though, there is a lack of hyperparameter understanding and best practices for EO – arguably due to severely less `graduate student descent' and benchmarking being performed for EO methods. Additionally, classical benchmarks from the evolutionary community provide few practical insights for Deep Learning applications. This poses challenges for newcomers to hardware-accelerated EO and hinders significant adoption. Hence, we establish a new benchmark of EO methods (NEB) tailored toward Deep Learning applications and exhaustively evaluate traditional and meta-learned EO. We investigate core scientific questions including resource allocation, fitness shaping, normalization, regularization & scalability of EO. The benchmark is open-sourced at https://github.com/neuroevobench/neuroevobench under Apache-2.0 license.

Language models have been successfully used to model natural signals, such as images, speech, and music. A key component of these models is a high quality neural compression model that can compress high-dimensional natural signals into lower dimensional discrete tokens. To that end, we introduce a high-fidelity universal neural audio compression algorithm that achieves ~90x compression of 44.1 KHz audio into tokens at just 8kbps bandwidth. We achieve this by combining advances in high-fidelity audio generation with better vector quantization techniques from the image domain, along with improved adversarial and reconstruction losses. We compress all domains (speech, environment, music, etc.) with a single universal model, making it widely applicable to generative modeling of all audio. We compare with competing audio compression algorithms, and find our method outperforms them significantly. We provide thorough ablations for every design choice, as well as open-source code and trained model weights. We hope our work can lay the foundation for the next generation of high-fidelity audio modeling.

We study gradient descent under linearly correlated noise. Our work is motivated by recent practical methods for optimization with differential privacy (DP), such as DP-FTRL, which achieve strong performance in settings where privacy amplification techniques are infeasible (such as in federated learning). These methods inject privacy noise through a matrix factorization mechanism, making the noise *linearly correlated* over iterations. We propose a simplified setting that distills key facets of these methods and isolates the impact of linearly correlated noise. We analyze the behavior of gradient descent in this setting, for both convex and non-convex functions. Our analysis is demonstrably tighter than prior work and recovers multiple important special cases exactly (including anticorrelated perturbed gradient descent). We use our results to develop new, effective matrix factorizations for differentially private optimization, and highlight the benefits of these factorizations theoretically and empirically.

Audio-visual learning has been a major pillar of multi-modal machine learning, where the community mostly focused on its $\textit{modality-aligned}$ setting, $\textit{i.e.}$, the audio and visual modality are $\textit{both}$ assumed to signal the prediction target. With the Look, Listen, and Parse dataset (LLP), we investigate the under-explored $\textit{unaligned}$ setting, where the goal is to recognize audio and visual events in a video with only weak labels observed. Such weak video-level labels only tell what events happen without knowing the modality they are perceived (audio, visual, or both). To enhance learning in this challenging setting, we incorporate large-scale contrastively pre-trained models as the modality teachers. A simple, effective, and generic method, termed $\textbf{V}$isual-$\textbf{A}$udio $\textbf{L}$abel Elab$\textbf{or}$ation (VALOR), is innovated to harvest modality labels for the training events. Empirical studies show that the harvested labels significantly improve an attentional baseline by $\textbf{8.0}$ in average F-score (Type@AV). Surprisingly, we found that modality-independent teachers outperform their modality-fused counterparts since they are noise-proof from the other potentially unaligned modality. Moreover, our best model achieves the new state-of-the-art on all metrics of LLP by a substantial margin ($\textbf{+5.4}$ F-score for Type@AV). VALOR is further generalized to Audio-Visual Event Localization and achieves the new state-of-the-art as well.

A room’s acoustic properties are a product of the room’s geometry, the objects within the room, and their specific positions. A room’s acoustic properties can be characterized by its impulse response (RIR) between a source and listener location, or roughly inferred from recordings of natural signals present in the room. Variations in the positions of objects in a room can effect measurable changes in the room’s acoustic properties, as characterized by the RIR. Existing datasets of RIRs either do not systematically vary positions of objects in an environment, or they consist of only simulated RIRs. We present SoundCam, the largest dataset of unique RIRs from in-the-wild rooms publicly released to date. It includes 5,000 10-channel real-world measurements of room impulse responses and 2,000 10-channel recordings of music in three different rooms, including a controlled acoustic lab, an in-the-wild living room, and a conference room, with different humans in positions throughout each room. We show that these measurements can be used for interesting tasks, such as detecting and identifying humans, and tracking their positions.

The stochastic block model (SBM) is a fundamental model for studying graph clustering or community detection in networks. It has received great attention in the last decade and the balanced case, i.e., assuming all clusters have large size, has been well studied. However, our understanding of SBM with unbalanced communities (arguably, more relevant in practice) is still limited. In this paper, we provide a simple SVD-based algorithm for recovering the communities in the SBM with communities of varying sizes. We improve upon a result of Ailon, Chen and Xu [ICML 2013; JMLR 2015] by removing the assumption that there is a large interval such that the sizes of clusters do not fall in, and also remove the dependency of the size of the recoverable clusters on the number of underlying clusters. We further complement our theoretical improvements with experimental comparisons. Under the planted clique conjecture, the size of the clusters that can be recovered by our algorithm is nearly optimal (up to poly-logarithmic factors) when the probability parameters are constant. As a byproduct, we obtain an efficient clustering algorithm with sublinear query complexity in a faulty oracle model, which is capable of detecting all clusters larger than $\tilde{\Omega}({\sqrt{n}})$, even in the presence of $\Omega(n)$ small clusters in the graph. In contrast, previous efficient algorithms that use a sublinear number of queries are incapable of recovering any large clusters if there are more than $\tilde{\Omega}(n^{2/5})$ small clusters.

Recent years have witnessed a surge of interest in integrating high-dimensional data captured by multisource sensors, driven by the impressive success of neural networks in integrating multimodal data. However, the integration of heterogeneous multimodal data poses a significant challenge, as confounding effects and dependencies among such heterogeneous data sources introduce unwanted variability and bias, leading to suboptimal performance of multimodal models. Therefore, it becomes crucial to normalize the low- or high-level features extracted from data modalities before their fusion takes place. This paper introduces RegBN, a novel approach for multimodal Batch Normalization with REGularization. RegBN uses the Frobenius norm as a regularizer term to address the side effects of confounders and underlying dependencies among different data sources. The proposed method generalizes well across multiple modalities and eliminates the need for learnable parameters, simplifying training and inference. We validate the effectiveness of RegBN on eight databases from five research areas, encompassing diverse modalities such as language, audio, image, video, depth, tabular, and 3D MRI. The proposed method demonstrates broad applicability across different architectures such as multilayer perceptrons, convolutional neural networks, and vision transformers, enabling effective normalization of both low- and high-level features in multimodal neural networks. RegBN is available at https://mogvision.github.io/RegBN.

We study a generalization of boosting to the multiclass setting. We introduce a weak learning condition for multiclass classification that captures the original notion of weak learnability as being “slightly better than random guessing”. We give a simple and efficient boosting algorithm, that does not require realizability assumptions and its sample and oracle complexity bounds are independent of the number of classes. In addition, we utilize our new boosting technique in several theoretical applications within the context of List PAC Learning. First, we establish an equivalence to weak PAC learning. Furthermore, we present a new result on boosting for list learners, as well as provide a novel proof for the characterization of multiclass PAC learning and List PAC learning. Notably, our technique gives rise to simplified algorithms and analysis compared to previous works.

In this paper, we present StyleTTS 2, a text-to-speech (TTS) model that leverages style diffusion and adversarial training with large speech language models (SLMs) to achieve human-level TTS synthesis. StyleTTS 2 differs from its predecessor by modeling styles as a latent random variable through diffusion models to generate the most suitable style for the text without requiring reference speech, achieving efficient latent diffusion while benefiting from the diverse speech synthesis offered by diffusion models. Furthermore, we employ large pre-trained SLMs, such as WavLM, as discriminators with our novel differentiable duration modeling for end-to-end training, resulting in improved speech naturalness. StyleTTS 2 surpasses human recordings on the single-speaker LJSpeech dataset and matches it on the multispeaker VCTK dataset as judged by native English speakers. Moreover, when trained on the LibriTTS dataset, our model outperforms previous publicly available models for zero-shot speaker adaptation. This work achieves the first human-level TTS on both single and multispeaker datasets, showcasing the potential of style diffusion and adversarial training with large SLMs. The audio demos and source code are available at https://styletts2.github.io/.

Correlation clustering is a fundamental optimization problem at the intersection of machine learning and theoretical computer science. Motivated by applications to big data processing, recent years have witnessed a flurry of results on this problem in the streaming model. In this model, the algorithm needs to process the input $n$-vertex graph by making one or few passes over the stream of its edges and using a limited memory, much smaller than the input size. All previous work on streaming correlation clustering have focused on semi-streaming algorithms with $\Omega(n)$ memory, whereas in this work, we study streaming algorithms with much smaller memory requirement of only $\text{polylog}{(n)}$ bits. This stringent memory requirement is in the same spirit of classical streaming algorithms that instead of recovering a full solution to the problem---which can be prohibitively large with such small memory as is the case in our problem---, aimed to learn certain statistical properties of their inputs. In our case, this translates to determining the ``(correlation) clusterability'' of input graphs, or more precisely, estimating the cost of the optimal correlation clustering solution. As our main result, we present two novel algorithms that in only $\text{polylog}{(n)}$ space are able to estimate the optimal correlation clustering cost up to some constant multiplicative factor plus some extra additive error. One of the algorithms outputs a $3$-multiplicative approximation plus $o(n^2)$ additive approximation, and the other one improves the additive error further down at the cost of increasing the multiplicative factor to some large constant. We then present new lower bounds that justify this mix of both multiplicative and additive error approximation in our algorithms.

A good supervised embedding for a specific machine learning task is only sensitive to changes in the label of interest and is invariant to other confounding factors. We leverage the concept of repeatability from measurement theory to describe this property and propose to use the intra-class correlation coefficient (ICC) to evaluate the repeatability of embeddings. We then propose a novel regularizer, the ICC regularizer, as a complementary component for contrastive losses to guide deep neural networks to produce embeddings with higher repeatability. We use simulated data to explain why the ICC regularizer works better on minimizing the intra-class variance than the contrastive loss alone. We implement the ICC regularizer and apply it to three speech tasks: speaker verification, voice style conversion, and a clinical application for detecting dysphonic voice. The experimental results demonstrate that adding an ICC regularizer can improve the repeatability of learned embeddings compared to only using the contrastive loss; further, these embeddings lead to improved performance in these downstream tasks.