Search

We extend some of our earlier results on the interconnection between ultrafilter extensions, and ultrapowers. Throughout we restrict ourselves to relational structures with one binary relation. Recently it was shown that for bounded structures, where a universal finite bound on the maximal in-and out-degree is given, ultrafilter extensions are elementary extensions of the original structures. Comparing the constructions, it seems that the real challenge is when the degree has no global finite bound, or there are elements with infinite degree. This is a first step towards this direction by slightly relaxing the notion of boundedness, called almost bounded structures. Among others, we show that members of this class are still elementary substructures of their extensions, elementary embeddings can be lifted up to the extensions, moreover for the countable case, the extensions are isomorphic to certain ultrapowers. We also comment on the modal logics they generate.

Let $G$ be a simple finite connected graph of order $n$ greater than or equal to $3$. We obtain the following results: (1). We apply a result of Hamada and Yoshimura from 1976 and some recent results of Alikhani and Soltani (2020) and Kalinowski and Pilsniak (2015) to determine the distinguishing chromatic number of the middle graph $M(G)$ of the graph $G$. In particular, the distinguishing chromatic number $\chi_{D}(M(G))$ of the middle graph $M(G)$ of the graph $G$ is $\Delta(G)+1$ except for four small graphs $C_{4}, K_{4}, C_{6}$, and $K_{3,3}$, and $\Delta(G)+2$ otherwise. (2). In 2016, Kalinowski, Pilsniak, and Wozniak introduced the total distinguishing number $D''(G)$ of $G$. Inspired by a recent result of Mirafzal (2024), we show that the distinguishing number $D(S(G))$ of the subdivision graph $S(G)$ of $G$ is $D''(G)$. Consequently, $D(S(G))$ is at most $\lceil \sqrt{\Delta(G)}\rceil$. (3). We obtain a sharp upper bound for the distinguishing chromatic number of the subdivision graph $S(G)$ of $G$ in terms of the distinguishing number of $G$., Comment: 8 pages, 4 figures

Multimodal Large Language Models (MLLMs) have tremendous potential to improve the accuracy, availability, and cost-effectiveness of healthcare by providing automated solutions or serving as aids to medical professionals. Despite promising first steps in developing medical MLLMs in the past few years, their capabilities and limitations are not well-understood. Recently, many benchmark datasets have been proposed that test the general medical knowledge of such models across a variety of medical areas. However, the systematic failure modes and vulnerabilities of such models are severely underexplored with most medical benchmarks failing to expose the shortcomings of existing models in this safety-critical domain. In this paper, we introduce MediConfusion, a challenging medical Visual Question Answering (VQA) benchmark dataset, that probes the failure modes of medical MLLMs from a vision perspective. We reveal that state-of-the-art models are easily confused by image pairs that are otherwise visually dissimilar and clearly distinct for medical experts. Strikingly, all available models (open-source or proprietary) achieve performance below random guessing on MediConfusion, raising serious concerns about the reliability of existing medical MLLMs for healthcare deployment. We also extract common patterns of model failure that may help the design of a new generation of more trustworthy and reliable MLLMs in healthcare., Comment: 17 Pages, 5 figures

This paper proposes a control technique for autonomous RC car racing. The presented method does not require any map-building phase beforehand since it operates only local path planning on the actual LiDAR point cloud. Racing control algorithms must have the capability to be optimized to the actual track layout for minimization of lap time. In the examined one, it is guaranteed with the improvement of the Stanley controller with additive control components to stabilize the movement in both low and high-speed ranges, and with the integration of an adaptive lookahead point to induce sharp and dynamic cornering for traveled distance reduction. The developed method is tested on a 1/10-sized RC car, and the tuning procedure from a base solution to the optimal setting in a real F1Tenth race is presented. Furthermore, the proposed method is evaluated with a comparison to a more simple reactive method, and in parallel to a more complex optimization-based technique that involves offline map building the global optimal trajectory calculation. The performance of the proposed method compared to the latter, referring to the lap time, is that the proposed one has only 8% lower average speed. This demonstrates that with appropriate tuning, a local planning-based method can be comparable with a more complex optimization-based one. Thus, the performance gap is lower than 10% from the state-of-the-art method. Moreover, the proposed technique has significantly higher similarity to real scenarios, therefore the results can be interesting in the context of automotive industry.

In recent years, the number and importance of autonomous racing leagues, and consequently the number of studies on them, has been growing. The seamless integration between different series has gained attention due to the scene's diversity. However, the high cost of full scale racing makes it a more accessible development model, to research at smaller form factors and scale up the achieved results. This paper presents a scalable architecture designed for autonomous racing that emphasizes modularity, adaptability to diverse configurations, and the ability to supervise parallel execution of pipelines that allows the use of different dynamic strategies. The system showcased consistent racing performance across different environments, demonstrated through successful participation in two relevant competitions. The results confirm the architecture's scalability and versatility, providing a robust foundation for the development of competitive autonomous racing systems. The successful application in real-world scenarios validates its practical effectiveness and highlights its potential for future advancements in autonomous racing technology.

In the past, analogues to Brooks' theorem have been found for various parameters of graph coloring for infinite locally finite connected graphs in ZFC. We prove these theorems are not provable in ZF (i.e. the Zermelo-Fraenkel set theory without the Axiom of Choice (AC)). Moreover, such theorems follow from Konig's Lemma (every infinite locally finite connected graph has a ray-a weak form of AC) in ZF. In ZF, we formulate new conditions for the existence of the distinguishing chromatic number, the distinguishing chromatic index, the total chromatic number, the total distinguishing chromatic number, the odd chromatic number, and the neighbor-distinguishing index in infinite locally finite connected graphs, which are equivalent to Konig's Lemma. In this direction, we strengthen a recent result of Stawiski from 2023. We also figured out the upper bound for list-distinguishing chromatic number for infinite graphs in ZFC (i.e. the Zermelo-Fraenkel set theory with the Axiom of Choice (AC))., Comment: Revised version

In a recent paper, Krawczyk proved that there are continuum many axiomatic extensions of global consequence associated with the modal system $E$ that do not admit the local deduction detachment theorem. In algebraic parlance, he showed that there are continuum many varieties of modal algebras lacking the congruence extension property. In this paper, we extend Krawczyk's results and construct a continuum of varieties of modal algebras that do not have the congruence extension property, but that do admit other, logically relevant properties, such as monotonicity, extensiveness, idempotency, normality, etc. This gives a continuum of axiomatic extensions of the corresponding modal systems not having the local deduction detachment theorem.

The main motivation of this paper is the study of first-order model theoretic properties of structures having their roots in modal logic. We will focus on the connections between ultrafilter extensions and ultrapowers. We show that certain structures (called bounded graphs) are elementary substructures of their ultrafilter extensions, moreover their modal logics coincide.

We prove analogs of Brooks' Theorem for the list-distinguishing chromatic number of different classes of simple finite connected graphs. Moreover, we determine two upper bounds for the list-distinguishing chromatic number of a graph G in terms of the coloring number of G and the list-chromatic number of G. We also determine the list-distinguishing chromatic number for various families of graphs (for example: the book graphs)., Comment: Revised version

The landscape of computational building blocks of efficient image restoration architectures is dominated by a combination of convolutional processing and various attention mechanisms. However, convolutional filters, while efficient, are inherently local and therefore struggle with modeling long-range dependencies in images. In contrast, attention excels at capturing global interactions between arbitrary image regions, but suffers from a quadratic cost in image dimension. In this work, we propose Serpent, an efficient architecture for high-resolution image restoration that combines recent advances in state space models (SSMs) with multi-scale signal processing in its core computational block. SSMs, originally introduced for sequence modeling, can maintain a global receptive field with a favorable linear scaling in input size. We propose a novel hierarchical architecture inspired by traditional signal processing principles, that converts the input image into a collection of sequences and processes them in a multi-scale fashion. Our experimental results demonstrate that Serpent can achieve reconstruction quality on par with state-of-the-art techniques, while requiring orders of magnitude less compute (up to $150$ fold reduction in FLOPS) and a factor of up to $5\times$ less GPU memory while maintaining a compact model size. The efficiency gains achieved by Serpent are especially notable at high image resolutions.

In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

We propose MusicRL, the first music generation system finetuned from human feedback. Appreciation of text-to-music models is particularly subjective since the concept of musicality as well as the specific intention behind a caption are user-dependent (e.g. a caption such as "upbeat work-out music" can map to a retro guitar solo or a techno pop beat). Not only this makes supervised training of such models challenging, but it also calls for integrating continuous human feedback in their post-deployment finetuning. MusicRL is a pretrained autoregressive MusicLM (Agostinelli et al., 2023) model of discrete audio tokens finetuned with reinforcement learning to maximise sequence-level rewards. We design reward functions related specifically to text-adherence and audio quality with the help from selected raters, and use those to finetune MusicLM into MusicRL-R. We deploy MusicLM to users and collect a substantial dataset comprising 300,000 pairwise preferences. Using Reinforcement Learning from Human Feedback (RLHF), we train MusicRL-U, the first text-to-music model that incorporates human feedback at scale. Human evaluations show that both MusicRL-R and MusicRL-U are preferred to the baseline. Ultimately, MusicRL-RU combines the two approaches and results in the best model according to human raters. Ablation studies shed light on the musical attributes influencing human preferences, indicating that text adherence and quality only account for a part of it. This underscores the prevalence of subjectivity in musical appreciation and calls for further involvement of human listeners in the finetuning of music generation models.

In set theory without the Axiom of Choice, we study the set-theoretic strength of a generalized version of the Rainbow Ramsey theorem and the Canonical Ramsey Theorem for pairs introduced by Erd\H{o}s and Rado, concerning their interrelation with several weak choice forms., Comment: Some new results were added

Deep learning models often suffer from forgetting previously learned information when trained on new data. This problem is exacerbated in federated learning (FL), where the data is distributed and can change independently for each user. Many solutions are proposed to resolve this catastrophic forgetting in a centralized setting. However, they do not apply directly to FL because of its unique complexities, such as privacy concerns and resource limitations. To overcome these challenges, this paper presents a framework for $\textbf{federated class incremental learning}$ that utilizes a generative model to synthesize samples from past distributions. This data can be later exploited alongside the training data to mitigate catastrophic forgetting. To preserve privacy, the generative model is trained on the server using data-free methods at the end of each task without requesting data from clients. Moreover, our solution does not demand the users to store old data or models, which gives them the freedom to join/leave the training at any time. Additionally, we introduce SuperImageNet, a new regrouping of the ImageNet dataset specifically tailored for federated continual learning. We demonstrate significant improvements compared to existing baselines through extensive experiments on multiple datasets., Comment: Accepted in NeurIPS 2023. arXiv admin note: text overlap with arXiv:2307.00497

In today's industrial challenges, it can be observed that the trends point in the direction of agile, wireless connected robots where elements of intelligence and control are implemented in the edge cloud. This paper outlines the roles of three key participants in the value chain of an industrial process: the network provider, the robot operator, and the customer. It proposes a scheme where the Quality of Service (QoS) parameters of the robot are fed into the network to inform network resource management. A sanding process use case is simulated to demonstrate the relationship between QoS and Quality of Experience (QoE) for each participant, quantitatively.

Due to deteriorating environmental conditions and increasing human activity, conservation efforts directed towards wildlife is crucial. Motion-activated camera traps constitute an efficient tool for tracking and monitoring wildlife populations across the globe. Supervised learning techniques have been successfully deployed to analyze such imagery, however training such techniques requires annotations from experts. Reducing the reliance on costly labelled data therefore has immense potential in developing large-scale wildlife tracking solutions with markedly less human labor. In this work we propose WildMatch, a novel zero-shot species classification framework that leverages multimodal foundation models. In particular, we instruction tune vision-language models to generate detailed visual descriptions of camera trap images using similar terminology to experts. Then, we match the generated caption to an external knowledge base of descriptions in order to determine the species in a zero-shot manner. We investigate techniques to build instruction tuning datasets for detailed animal description generation and propose a novel knowledge augmentation technique to enhance caption quality. We demonstrate the performance of WildMatch on a new camera trap dataset collected in the Magdalena Medio region of Colombia., Comment: 18 pages, 9 figures

We work with simple graphs in ZF (Zermelo--Fraenkel set theory without the Axiom of Choice (AC)) and assume that the sets of colors can be either well-orderable or non-well-orderable to prove that the following statements are equivalent to K\H{o}nig Lemma: (a) Any infinite locally finite connected graph G such that the minimum degree of G is greater than k, has a chromatic number for any fixed integer k greater than or equal to 2. (b) Any infinite locally finite connected graph has a chromatic index. (c) Any infinite locally finite connected graph has a distinguishing number. (d) Any infinite locally finite connected graph has a distinguishing index. Our results strengthen some results of Stawiski from a recent paper on the role of the Axiom of Choice in proper and distinguishing colorings since he assumed that the sets of colors can be well-ordered. We also formulate new conditions for the existence of irreducible proper coloring, minimal edge cover, maximal matching, and minimal dominating set in connected bipartite graphs and locally finite connected graphs, which are either equivalent to AC or K\H{o}nig Lemma. Moreover, we show that if the Axiom of Choice for families of 2 element sets holds, then the Shelah--Soifer graph has a minimal dominating set., Comment: This revised version is accepted in a journal

Inverse problems arise in a multitude of applications, where the goal is to recover a clean signal from noisy and possibly (non)linear observations. The difficulty of a reconstruction problem depends on multiple factors, such as the ground truth signal structure, the severity of the degradation and the complex interactions between the above. This results in natural sample-by-sample variation in the difficulty of a reconstruction problem. Our key observation is that most existing inverse problem solvers lack the ability to adapt their compute power to the difficulty of the reconstruction task, resulting in subpar performance and wasteful resource allocation. We propose a novel method, $\textit{severity encoding}$, to estimate the degradation severity of corrupted signals in the latent space of an autoencoder. We show that the estimated severity has strong correlation with the true corruption level and can provide useful hints on the difficulty of reconstruction problems on a sample-by-sample basis. Furthermore, we propose a reconstruction method based on latent diffusion models that leverages the predicted degradation severities to fine-tune the reverse diffusion sampling trajectory and thus achieve sample-adaptive inference times. Our framework, Flash-Diffusion, acts as a wrapper that can be combined with any latent diffusion-based baseline solver, imbuing it with sample-adaptivity and acceleration. We perform experiments on both linear and nonlinear inverse problems and demonstrate that our technique greatly improves the performance of the baseline solver and achieves up to $10\times$ acceleration in mean sampling speed., Comment: 30 pages, 21 figures, published at the 41st International Conference on Machine Learning, Vienna, Austria, 2024

We derive the basic postulates of quantum physics from a few very simple operational assumptions based exclusively on the relative frequencies of observable events (measurement operations and measurement outcomes). We isolate a notion which can be identified with the system's own state, in the sense that it characterizes the system's probabilistic behavior against all possible measurement operations. We investigate some important features of the possible states of the system. All those investigations remain within the framework of classical Kolmogorovian probability theory, meaning that any physical system (traditionally categorized as classical or quantum) that can be described in operational terms can be described within classical Kolmogorovian probability theory. In the second part of the paper we show that anything that can be described in operational terms can, if we wish, be represented in the Hilbert space quantum mechanical formalism. The outcomes of each measurement can be represented by a system of pairwise orthogonal closed subspaces spanning the entire Hilbert space; the states of the system can be represented by pure state operators, and the probabilities of the outcomes can be reproduced by the usual trace formula. Each real valued quantity can be associated with a suitable self-adjoint operator, such that the possible measurement results are the eigenvalues and the outcome events are represented by the eigenspaces, according to the spectral decomposition of the operator in question. This suggests that the basic postulates of quantum theory are in fact analytic statements: they do not tell us anything about a physical system beyond the fact that the system can be described in operational terms. This is almost true. At the end of the paper we discuss a few subtle points where the representation we obtained is not completely identical with standard quantum mechanics., Comment: 40 pages

We present TokenSplit, a speech separation model that acts on discrete token sequences. The model is trained on multiple tasks simultaneously: separate and transcribe each speech source, and generate speech from text. The model operates on transcripts and audio token sequences and achieves multiple tasks through masking of inputs. The model is a sequence-to-sequence encoder-decoder model that uses the Transformer architecture. We also present a "refinement" version of the model that predicts enhanced audio tokens from the audio tokens of speech separated by a conventional separation model. Using both objective metrics and subjective MUSHRA listening tests, we show that our model achieves excellent performance in terms of separation, both with or without transcript conditioning. We also measure the automatic speech recognition (ASR) performance and provide audio samples of speech synthesis to demonstrate the additional utility of our model., Comment: INTERSPEECH 2023, project webpage with audio demos at https://google-research.github.io/sound-separation/papers/tokensplit

Quasi-Newton methods still face significant challenges in training large-scale neural networks due to additional compute costs in the Hessian related computations and instability issues in stochastic training. A well-known method, L-BFGS that efficiently approximates the Hessian using history parameter and gradient changes, suffers convergence instability in stochastic training. So far, attempts that adapt L-BFGS to large-scale stochastic training incur considerable extra overhead, which offsets its convergence benefits in wall-clock time. In this paper, we propose mL-BFGS, a lightweight momentum-based L-BFGS algorithm that paves the way for quasi-Newton (QN) methods in large-scale distributed deep neural network (DNN) optimization. mL-BFGS introduces a nearly cost-free momentum scheme into L-BFGS update and greatly reduces stochastic noise in the Hessian, therefore stabilizing convergence during stochastic optimization. For model training at a large scale, mL-BFGS approximates a block-wise Hessian, thus enabling distributing compute and memory costs across all computing nodes. We provide a supporting convergence analysis for mL-BFGS in stochastic settings. To investigate mL-BFGS potential in large-scale DNN training, we train benchmark neural models using mL-BFGS and compare performance with baselines (SGD, Adam, and other quasi-Newton methods). Results show that mL-BFGS achieves both noticeable iteration-wise and wall-clock speedup., Comment: Accepted to TMLR 2023 (21 pages, 8 figures)

Deep learning models are prone to forgetting information learned in the past when trained on new data. This problem becomes even more pronounced in the context of federated learning (FL), where data is decentralized and subject to independent changes for each user. Continual Learning (CL) studies this so-called \textit{catastrophic forgetting} phenomenon primarily in centralized settings, where the learner has direct access to the complete training dataset. However, applying CL techniques to FL is not straightforward due to privacy concerns and resource limitations. This paper presents a framework for federated class incremental learning that utilizes a generative model to synthesize samples from past distributions instead of storing part of past data. Then, clients can leverage the generative model to mitigate catastrophic forgetting locally. The generative model is trained on the server using data-free methods at the end of each task without requesting data from clients. Therefore, it reduces the risk of data leakage as opposed to training it on the client's private data. We demonstrate significant improvements for the CIFAR-100 dataset compared to existing baselines.

We introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation. AudioPaLM inherits the capability to preserve paralinguistic information such as speaker identity and intonation from AudioLM and the linguistic knowledge present only in text large language models such as PaLM-2. We demonstrate that initializing AudioPaLM with the weights of a text-only large language model improves speech processing, successfully leveraging the larger quantity of text training data used in pretraining to assist with the speech tasks. The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training. AudioPaLM also demonstrates features of audio language models, such as transferring a voice across languages based on a short spoken prompt. We release examples of our method at https://google-research.github.io/seanet/audiopalm/examples, Comment: Technical report

Subobject independence as morphism co-possibility has recently been defined in [2] and studied in the context of algebraic quantum field theory. This notion of independence is handy when it comes to systems coming from physics, but when directly applied to classical algebras, subobject independence is not entirely satisfactory. The sole purpose of this note is to introduce the notion of subalgebra independence, which is a slight variation of subobject independence, yet this modification enables us to connect subalgebra independence to more traditional notions of independence. Apart from drawing connections between subalgebra independence and coproducts and congruences, we mainly illustrate the notion by discussing examples.

We present SoundStorm, a model for efficient, non-autoregressive audio generation. SoundStorm receives as input the semantic tokens of AudioLM, and relies on bidirectional attention and confidence-based parallel decoding to generate the tokens of a neural audio codec. Compared to the autoregressive generation approach of AudioLM, our model produces audio of the same quality and with higher consistency in voice and acoustic conditions, while being two orders of magnitude faster. SoundStorm generates 30 seconds of audio in 0.5 seconds on a TPU-v4. We demonstrate the ability of our model to scale audio generation to longer sequences by synthesizing high-quality, natural dialogue segments, given a transcript annotated with speaker turns and a short prompt with the speakers' voices.

Diffusion models have established new state of the art in a multitude of computer vision tasks, including image restoration. Diffusion-based inverse problem solvers generate reconstructions of exceptional visual quality from heavily corrupted measurements. However, in what is widely known as the perception-distortion trade-off, the price of perceptually appealing reconstructions is often paid in declined distortion metrics, such as PSNR. Distortion metrics measure faithfulness to the observation, a crucial requirement in inverse problems. In this work, we propose a novel framework for inverse problem solving, namely we assume that the observation comes from a stochastic degradation process that gradually degrades and noises the original clean image. We learn to reverse the degradation process in order to recover the clean image. Our technique maintains consistency with the original measurement throughout the reverse process, and allows for great flexibility in trading off perceptual quality for improved distortion metrics and sampling speedup via early-stopping. We demonstrate the efficiency of our method on different high-resolution datasets and inverse problems, achieving great improvements over other state-of-the-art diffusion-based methods with respect to both perceptual and distortion metrics., Comment: 30 pages, 15 figures, published at the 41st International Conference on Machine Learning, Vienna, Austria, 2024

We introduce LMCodec, a causal neural speech codec that provides high quality audio at very low bitrates. The backbone of the system is a causal convolutional codec that encodes audio into a hierarchy of coarse-to-fine tokens using residual vector quantization. LMCodec trains a Transformer language model to predict the fine tokens from the coarse ones in a generative fashion, allowing for the transmission of fewer codes. A second Transformer predicts the uncertainty of the next codes given the past transmitted codes, and is used to perform conditional entropy coding. A MUSHRA subjective test was conducted and shows that the quality is comparable to reference codecs at higher bitrates. Example audio is available at https://mjenrungrot.github.io/chrome-media-audio-papers/publications/lmcodec., Comment: 5 pages, accepted to ICASSP 2023, project page: https://mjenrungrot.github.io/chrome-media-audio-papers/publications/lmcodec

We introduce SPEAR-TTS, a multi-speaker text-to-speech (TTS) system that can be trained with minimal supervision. By combining two types of discrete speech representations, we cast TTS as a composition of two sequence-to-sequence tasks: from text to high-level semantic tokens (akin to "reading") and from semantic tokens to low-level acoustic tokens ("speaking"). Decoupling these two tasks enables training of the "speaking" module using abundant audio-only data, and unlocks the highly efficient combination of pretraining and backtranslation to reduce the need for parallel data when training the "reading" component. To control the speaker identity, we adopt example prompting, which allows SPEAR-TTS to generalize to unseen speakers using only a short sample of 3 seconds, without any explicit speaker representation or speaker-id labels. Our experiments demonstrate that SPEAR-TTS achieves a character error rate that is competitive with state-of-the-art methods using only 15 minutes of parallel data, while matching ground-truth speech in terms of naturalness and acoustic quality, as measured in subjective tests.

We introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption. To support future research, we publicly release MusicCaps, a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts., Comment: Supplementary material at https://google-research.github.io/seanet/musiclm/examples and https://kaggle.com/datasets/googleai/musiccaps

We introduce AudioLM, a framework for high-quality audio generation with long-term consistency. AudioLM maps the input audio to a sequence of discrete tokens and casts audio generation as a language modeling task in this representation space. We show how existing audio tokenizers provide different trade-offs between reconstruction quality and long-term structure, and we propose a hybrid tokenization scheme to achieve both objectives. Namely, we leverage the discretized activations of a masked language model pre-trained on audio to capture long-term structure and the discrete codes produced by a neural audio codec to achieve high-quality synthesis. By training on large corpora of raw audio waveforms, AudioLM learns to generate natural and coherent continuations given short prompts. When trained on speech, and without any transcript or annotation, AudioLM generates syntactically and semantically plausible speech continuations while also maintaining speaker identity and prosody for unseen speakers. Furthermore, we demonstrate how our approach extends beyond speech by generating coherent piano music continuations, despite being trained without any symbolic representation of music.

We present a method to separate speech signals from noisy environments in the embedding space of a neural audio codec. We introduce a new training procedure that allows our model to produce structured encodings of audio waveforms given by embedding vectors, where one part of the embedding vector represents the speech signal, and the rest represent the environment. We achieve this by partitioning the embeddings of different input waveforms and training the model to faithfully reconstruct audio from mixed partitions, thereby ensuring each partition encodes a separate audio attribute. As use cases, we demonstrate the separation of speech from background noise or from reverberation characteristics. Our method also allows for targeted adjustments of the audio output characteristics., Comment: Accepted at ICASSP 2023

We overview angle-based models to study planar symmetric central configurations of four bodies. We present models to determine the masses given the shape of the configuration and the shape-type of the configuration given the masses. We also describe a diagram-based method for the counting of the shape-types in the case of the concave configurations. As an application, we determine planar symmetric central configurations containing bodies of Earth's and Moon's masses., Comment: 15 pages, submitted to Physical Review E

In accelerated MRI reconstruction, the anatomy of a patient is recovered from a set of under-sampled and noisy measurements. Deep learning approaches have been proven to be successful in solving this ill-posed inverse problem and are capable of producing very high quality reconstructions. However, current architectures heavily rely on convolutions, that are content-independent and have difficulties modeling long-range dependencies in images. Recently, Transformers, the workhorse of contemporary natural language processing, have emerged as powerful building blocks for a multitude of vision tasks. These models split input images into non-overlapping patches, embed the patches into lower-dimensional tokens and utilize a self-attention mechanism that does not suffer from the aforementioned weaknesses of convolutional architectures. However, Transformers incur extremely high compute and memory cost when 1) the input image resolution is high and 2) when the image needs to be split into a large number of patches to preserve fine detail information, both of which are typical in low-level vision problems such as MRI reconstruction, having a compounding effect. To tackle these challenges, we propose HUMUS-Net, a hybrid architecture that combines the beneficial implicit bias and efficiency of convolutions with the power of Transformer blocks in an unrolled and multi-scale network. HUMUS-Net extracts high-resolution features via convolutional blocks and refines low-resolution features via a novel Transformer-based multi-scale feature extractor. Features from both levels are then synthesized into a high-resolution output reconstruction. Our network establishes new state of the art on the largest publicly available MRI dataset, the fastMRI dataset. We further demonstrate the performance of HUMUS-Net on two other popular MRI datasets and perform fine-grained ablation studies to validate our design., Comment: 18 pages, 11 figures, NeurIPS 2022

We propose SpeechPainter, a model for filling in gaps of up to one second in speech samples by leveraging an auxiliary textual input. We demonstrate that the model performs speech inpainting with the appropriate content, while maintaining speaker identity, prosody and recording environment conditions, and generalizing to unseen speakers. Our approach significantly outperforms baselines constructed using adaptive TTS, as judged by human raters in side-by-side preference and MOS tests., Comment: Submitted to Interspeech 2022

Greedy navigation/routing plays an important role in geometric routing of networks because of its locality and simplicity. This can operate in geometrically embedded networks in a distributed manner, distances are calculated based on coordinates of network nodes for choosing the next hop in the routing. Based only on node coordinates in any metric space, the Greedy Navigational Core (GNC) can be identi ?ed as the minimum set of links between these nodes which provides 100% greedy navigability. In this paper we perform results on structural greedy navigability as the level of presence of Greedy Navigational Cores in structural networks of the Human Brain.

The increasing availability of massive data sets poses a series of challenges for machine learning. Prominent among these is the need to learn models under hardware or human resource constraints. In such resource-constrained settings, a simple yet powerful approach is to operate on small subsets of the data. Coresets are weighted subsets of the data that provide approximation guarantees for the optimization objective. However, existing coreset constructions are highly model-specific and are limited to simple models such as linear regression, logistic regression, and $k$-means. In this work, we propose a generic coreset construction framework that formulates the coreset selection as a cardinality-constrained bilevel optimization problem. In contrast to existing approaches, our framework does not require model-specific adaptations and applies to any twice differentiable model, including neural networks. We show the effectiveness of our framework for a wide range of models in various settings, including training non-convex models online and batch active learning.

Deep neural networks have emerged as very successful tools for image restoration and reconstruction tasks. These networks are often trained end-to-end to directly reconstruct an image from a noisy or corrupted measurement of that image. To achieve state-of-the-art performance, training on large and diverse sets of images is considered critical. However, it is often difficult and/or expensive to collect large amounts of training images. Inspired by the success of Data Augmentation (DA) for classification problems, in this paper, we propose a pipeline for data augmentation for accelerated MRI reconstruction and study its effectiveness at reducing the required training data in a variety of settings. Our DA pipeline, MRAugment, is specifically designed to utilize the invariances present in medical imaging measurements as naive DA strategies that neglect the physics of the problem fail. Through extensive studies on multiple datasets we demonstrate that in the low-data regime DA prevents overfitting and can match or even surpass the state of the art while using significantly fewer training data, whereas in the high-data regime it has diminishing returns. Furthermore, our findings show that DA can improve the robustness of the model against various shifts in the test distribution., Comment: 27 pages, 19 figures, to be published in ICML2021

A crucial aspect for the successful deployment of audio-based models "in-the-wild" is the robustness to the transformations introduced by heterogeneous acquisition conditions. In this work, we propose a method to perform one-shot microphone style transfer. Given only a few seconds of audio recorded by a target device, MicAugment identifies the transformations associated to the input acquisition pipeline and uses the learned transformations to synthesize audio as if it were recorded under the same conditions as the target audio. We show that our method can successfully apply the style transfer to real audio and that it significantly increases model robustness when used as data augmentation in the downstream tasks.

Active learning is an effective technique for reducing the labeling cost by improving data efficiency. In this work, we propose a novel batch acquisition strategy for active learning in the setting where the model training is performed in a semi-supervised manner. We formulate our approach as a data summarization problem via bilevel optimization, where the queried batch consists of the points that best summarize the unlabeled data pool. We show that our method is highly effective in keyword detection tasks in the regime when only few labeled samples are available.

Transfer learning has emerged as a powerful technique for improving the performance of machine learning models on new domains where labeled training data may be scarce. In this approach a model trained for a source task, where plenty of labeled training data is available, is used as a starting point for training a model on a related target task with only few labeled training data. Despite recent empirical success of transfer learning approaches, the benefits and fundamental limits of transfer learning are poorly understood. In this paper we develop a statistical minimax framework to characterize the fundamental limits of transfer learning in the context of regression with linear and one-hidden layer neural network models. Specifically, we derive a lower-bound for the target generalization error achievable by any algorithm as a function of the number of labeled source and target data as well as appropriate notions of similarity between the source and target tasks. Our lower bound provides new insights into the benefits and limitations of transfer learning. We further corroborate our theoretical finding with various experiments.

Coresets are small data summaries that are sufficient for model training. They can be maintained online, enabling efficient handling of large data streams under resource constraints. However, existing constructions are limited to simple models such as k-means and logistic regression. In this work, we propose a novel coreset construction via cardinality-constrained bilevel optimization. We show how our framework can efficiently generate coresets for deep neural networks, and demonstrate its empirical benefits in continual learning and in streaming settings., Comment: NeurIPS 2020