arXiv subjects

Arun Babu

Publications and source records attributed to Arun Babu.

14 recordsLinked to original sources

arXiv2024

Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model

We introduce Transfusion, a recipe for training a multi-modal model over discrete and continuous data. Transfusion combines the language modeling loss function (next token prediction) with diffusion to train a single transformer over mixed-modality sequences. We pretrain multiple Transfusion models up to 7B parameters from scratch on a mixture of text and image data, establishing scaling laws with respect to a variety of uni- and cross-modal benchmarks. Our experiments show that Transfusion scales significantly better than quantizing images and training a language model over discrete image tokens. By introducing modality-specific encoding and decoding layers, we can further improve the performance of Transfusion models, and even compress each image to just 16 patches. We further demonstrate that scaling our Transfusion recipe to 7B parameters and 2T multi-modal tokens produces a model that can generate images and text on a par with similar scale diffusion models and language models, reaping the benefits of both worlds.

cs.AI↗

arXiv2023

Toward Joint Language Modeling for Speech Units and Text

Speech and text are two major forms of human language. The research community has been focusing on mapping speech to text or vice versa for many years. However, in the field of language modeling, very little effort has been made to model them jointly. In light of this, we explore joint language modeling for speech units and text. Specifically, we compare different speech tokenizers to transform continuous speech signals into discrete units and use different methods to construct mixed speech-text data. We introduce automatic metrics to evaluate how well the joint LM mixes speech and text. We also fine-tune the LM on downstream spoken language understanding (SLU) tasks with different modalities (speech or text) and test its performance to assess the model's learning of shared representations. Our results show that by mixing speech units and text with our proposed mixing techniques, the joint LM improves over a speech-only baseline on SLU tasks and shows zero-shot cross-modal transferability.

cs.CL↗

arXiv2023

Scaling Autoregressive Multi-Modal Models: Pretraining and Instruction Tuning

We present CM3Leon (pronounced "Chameleon"), a retrieval-augmented, token-based, decoder-only multi-modal language model capable of generating and infilling both text and images. CM3Leon uses the CM3 multi-modal architecture but additionally shows the extreme benefits of scaling up and tuning on more diverse instruction-style data. It is the first multi-modal model trained with a recipe adapted from text-only language models, including a large-scale retrieval-augmented pre-training stage and a second multi-task supervised fine-tuning (SFT) stage. It is also a general-purpose model that can do both text-to-image and image-to-text generation, allowing us to introduce self-contained contrastive decoding methods that produce high-quality outputs. Extensive experiments demonstrate that this recipe is highly effective for multi-modal models. CM3Leon achieves state-of-the-art performance in text-to-image generation with 5x less training compute than comparable methods (zero-shot MS-COCO FID of 4.88). After SFT, CM3Leon can also demonstrate unprecedented levels of controllability in tasks ranging from language-guided image editing to image-controlled generation and segmentation.

cs.LG↗

arXiv2023

Efficient Self-supervised Learning with Contextualized Target Representations for Vision, Speech and Language

Current self-supervised learning algorithms are often modality-specific and require large amounts of computational resources. To address these issues, we increase the training efficiency of data2vec, a learning objective that generalizes across several modalities. We do not encode masked tokens, use a fast convolutional decoder and amortize the effort to build teacher representations. data2vec 2.0 benefits from the rich contextualized target representations introduced in data2vec which enable a fast self-supervised learner. Experiments on ImageNet-1K image classification show that data2vec 2.0 matches the accuracy of Masked Autoencoders in 16.4x lower pre-training time, on Librispeech speech recognition it performs as well as wav2vec 2.0 in 10.6x less time, and on GLUE natural language understanding it matches a retrained RoBERTa model in half the time. Trading some speed for accuracy results in ImageNet-1K top-1 accuracy of 86.8\% with a ViT-L model trained for 150 epochs.

cs.LG↗

arXiv2023

Scaling Speech Technology to 1,000+ Languages

Expanding the language coverage of speech technology has the potential to improve access to information for many more people. However, current speech technology is restricted to about one hundred languages which is a small fraction of the over 7,000 languages spoken around the world. The Massively Multilingual Speech (MMS) project increases the number of supported languages by 10-40x, depending on the task. The main ingredients are a new dataset based on readings of publicly available religious texts and effectively leveraging self-supervised learning. We built pre-trained wav2vec 2.0 models covering 1,406 languages, a single multilingual automatic speech recognition model for 1,107 languages, speech synthesis models for the same number of languages, as well as a language identification model for 4,017 languages. Experiments show that our multilingual speech recognition model more than halves the word error rate of Whisper on 54 languages of the FLEURS benchmark while being trained on a small fraction of the labeled data.

cs.CL↗

arXiv2022

data2vec: A General Framework for Self-supervised Learning in Speech, Vision and Language

While the general idea of self-supervised learning is identical across modalities, the actual algorithms and objectives differ widely because they were developed with a single modality in mind. To get us closer to general self-supervised learning, we present data2vec, a framework that uses the same learning method for either speech, NLP or computer vision. The core idea is to predict latent representations of the full input data based on a masked view of the input in a self-distillation setup using a standard Transformer architecture. Instead of predicting modality-specific targets such as words, visual tokens or units of human speech which are local in nature, data2vec predicts contextualized latent representations that contain information from the entire input. Experiments on the major benchmarks of speech recognition, image classification, and natural language understanding demonstrate a new state of the art or competitive performance to predominant approaches.

cs.LG↗

arXiv2022

Proof of Backhaul: Trustfree Measurement of Broadband Bandwidth

Recent years have seen the emergence of decentralized wireless networks consisting of nodes hosted by many individuals and small enterprises, reawakening the decades-old dream of open networking. These networks have been deployed in an organic, distributed manner and are driven by new economic models resting on tokenized incentives. A critical requirement for the incentives to scale is the ability to prove network performance in a decentralized trustfree manner, i.e., a Byzantine fault tolerant network telemetry system. In this paper, we present a Proof of Backhaul (PoB) protocol which measures the bandwidth of the (broadband) backhaul link of a wireless access point, termed prover, in a decentralized and trustfree manner. In particular, our proposed protocol is the first one to satisfy the following two properties: (1) Trustfree. Bandwidth measurement is secure against Byzantine attacks by collaborations of challenge servers and the prover. (2) Open. The barrier-to-entry for being a challenge server is low; there is no requirement of having a low latency and high throughput path to the measured link. At a high-level, our protocol aggregates the challenge traffic from multiple challenge servers and uses cryptographic primitives to ensure that a subset of challengers or, even challengers and provers, cannot maliciously modify results in their favor. A formal security model allows us to establish guarantees of accurate bandwidth measurement as a function of the fraction of malicious actors. Our evaluation shows that our PoB protocol can verify backhaul bandwidth of up to 1000 Mbps with less than 8% error using measurements lasting only 100 ms. The measurement accuracy is not affected in the presence of corrupted challengers. Importantly, the basic verification protocol lends itself to a minor modification that can measure available bandwidth even in the presence of cross-traffic.

cs.CR↗

arXiv2021

XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale

This paper presents XLS-R, a large-scale model for cross-lingual speech representation learning based on wav2vec 2.0. We train models with up to 2B parameters on nearly half a million hours of publicly available speech audio in 128 languages, an order of magnitude more public data than the largest known prior work. Our evaluation covers a wide range of tasks, domains, data regimes and languages, both high and low-resource. On the CoVoST-2 speech translation benchmark, we improve the previous state of the art by an average of 7.4 BLEU over 21 translation directions into English. For speech recognition, XLS-R improves over the best known prior work on BABEL, MLS, CommonVoice as well as VoxPopuli, lowering error rates by 14-34% relative on average. XLS-R also sets a new state of the art on VoxLingua107 language identification. Moreover, we show that with sufficient model size, cross-lingual pretraining can outperform English-only pretraining when translating English speech into other languages, a setting which favors monolingual pretraining. We hope XLS-R can help to improve speech processing tasks for many more languages of the world.

cs.CL↗

arXiv2021

Span Pointer Networks for Non-Autoregressive Task-Oriented Semantic Parsing

An effective recipe for building seq2seq, non-autoregressive, task-oriented parsers to map utterances to semantic frames proceeds in three steps: encoding an utterance $x$, predicting a frame's length |y|, and decoding a |y|-sized frame with utterance and ontology tokens. Though empirically strong, these models are typically bottlenecked by length prediction, as even small inaccuracies change the syntactic and semantic characteristics of resulting frames. In our work, we propose span pointer networks, non-autoregressive parsers which shift the decoding task from text generation to span prediction; that is, when imputing utterance spans into frame slots, our model produces endpoints (e.g., [i, j]) as opposed to text (e.g., "6pm"). This natural quantization of the output space reduces the variability of gold frames, therefore improving length prediction and, ultimately, exact match. Furthermore, length prediction is now responsible for frame syntax and the decoder is responsible for frame semantics, resulting in a coarse-to-fine model. We evaluate our approach on several task-oriented semantic parsing datasets. Notably, we bridge the quality gap between non-autogressive and autoregressive parsers, achieving 87 EM on TOPv2 (Chen et al. 2020). Furthermore, due to our more consistent gold frames, we show strong improvements in model generalization in both cross-domain and cross-lingual transfer in low-resource settings. Finally, due to our diminished output vocabulary, we observe 70% reduction in latency and 83% reduction in memory at beam size 5 compared to prior non-autoregressive parsers.

cs.CL↗

arXiv2021

Latency-Aware Neural Architecture Search with Multi-Objective Bayesian Optimization

When tuning the architecture and hyperparameters of large machine learning models for on-device deployment, it is desirable to understand the optimal trade-offs between on-device latency and model accuracy. In this work, we leverage recent methodological advances in Bayesian optimization over high-dimensional search spaces and multi-objective Bayesian optimization to efficiently explore these trade-offs for a production-scale on-device natural language understanding model at Facebook.

cs.LG↗

arXiv2021

Non-Autoregressive Semantic Parsing for Compositional Task-Oriented Dialog

Semantic parsing using sequence-to-sequence models allows parsing of deeper representations compared to traditional word tagging based models. In spite of these advantages, widespread adoption of these models for real-time conversational use cases has been stymied by higher compute requirements and thus higher latency. In this work, we propose a non-autoregressive approach to predict semantic parse trees with an efficient seq2seq model architecture. By combining non-autoregressive prediction with convolutional neural networks, we achieve significant latency gains and parameter size reduction compared to traditional RNN models. Our novel architecture achieves up to an 81% reduction in latency on TOP dataset and retains competitive performance to non-pretrained models on three different semantic parsing datasets. Our code is available at https://github.com/facebookresearch/pytext

cs.CL↗

arXiv2020

Lightweight Convolutional Representations for On-Device Natural Language Processing

The increasing computational and memory complexities of deep neural networks have made it difficult to deploy them on low-resource electronic devices (e.g., mobile phones, tablets, wearables). Practitioners have developed numerous model compression methods to address these concerns, but few have condensed input representations themselves. In this work, we propose a fast, accurate, and lightweight convolutional representation that can be swapped into any neural model and compressed significantly (up to 32x) with a negligible reduction in performance. In addition, we show gains over recurrent representations when considering resource-centric metrics (e.g., model file size, latency, memory usage) on a Samsung Galaxy S9.

cs.CL↗

arXiv2014

Coronal Mass Ejections from the Sun - Propagation and Near Earth Effects

Owing to our dependance on spaceborne technology, an awareness of disturbances in the near-Earth space environment is proving to be increasingly crucial. Earth-directed Coronal mass ejections (CMEs) emanating from the Sun are the primary drivers of space weather disturbances. Studies of CMEs, their kinematics, and their near-Earth effects are therefore gaining in importance. The effect of CMEs near the Earth is often manifested as transient decreases in galactic cosmic ray intensity, which are called Forbush decreases (FDs). In this thesis we probe the structure of CMEs and their associated shocks using FD observations by the GRAPES-3 muon telescope at Ooty. We have established that the cumulative diffusion of galactic cosmic rays into the CME is the dominant mechanism for causing FDs (Chapter 3). This diffusion takes place through a turbulent sheath region between the CME and the shock. One of our main results concerns the turbulence level in this region. We have quantitatively established that cross-field diffusion aided by magnetic field turbulence accounts for the observed lag between the FD and the magnetic field enhancement of the sheath region (Chapter 4 ). We have also investigated the nature of the driving forces acting on CMEs in this thesis. Using CME data from the SECCHI coronagraphs aboard STEREO sapcecraft, we have found evidence for the non-force-free nature of the magnetic field configuration inside these CMEs, which is the basis for the (often-invoked) Lorentz self-force driving (Chapter 5). Taken together the work presented in this thesis is a comprehensive attempt to characterise CME propagation from typical coronagraph fields of view to the Earth.

astro-ph.SR↗

arXiv2013

High rigidity Forbush decreases: due to CMEs or shocks?

We seek to identify the primary agents causing Forbush decreases (FDs) observed at the Earth in high rigidity cosmic rays. In particular, we ask if such FDs are caused mainly by coronal mass ejections (CMEs) from the Sun that are directed towards the Earth, or by their associated shocks. We use the muon data at cutoff rigidities ranging from 14 to 24 GV from the GRAPES-3 tracking muon telescope to identify FD events. We select those FD events that have a reasonably clean profile, and can be reasonably well associated with an Earth-directed CME and its associated shock. We employ two models: one that considers the CME as the sole cause of the FD (the CME-only model) and one that considers the shock as the only agent causing the FD (the shock-only model). We use an extensive set of observationally determined parameters for both these models. The only free parameter in these models is the level of MHD turbulence in the sheath region, which mediates cosmic ray diffusion (into the CME, for the CME-only model and across the shock sheath, for the shock-only model). We find that good fits to the GRAPES-3 multi-rigidity data using the CME-only model require turbulence levels in the CME sheath region that are only slightly higher than those estimated for the quiet solar wind. On the other hand, reasonable model fits with the shock-only model require turbulence levels in the sheath region that are an order of magnitude higher than those in the quiet solar wind. This observation naturally leads to the conclusion that the Earth-directed CMEs are the primary contributors to FDs observed in high rigidity cosmic rays.

astro-ph.SR↗