Jump to Content
Dustin Tran

Dustin Tran

I am a research scientist at Google Brain. I am broadly interested in advancing science and intelligence, and where the ideas involve probability, programs, and/or neural nets. URL: dustintran.com

Research Areas

Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, desc
  • Year
  • Year, desc
    Preview abstract The scaling of Transformers has driven breakthrough capabilities for language models. At present, the largest large language models (LLMs) contain upwards of 100B parameters. Vision Transformers (ViT) have introduced the same architecture to image and video modeling, but these have not yet been successfully scaled to nearly the same degree; the largest dense ViT contains 4B parameters. We present a recipe for highly efficient training of a 22B-parameter ViT and perform a wide variety of experiments on the resulting model. When evaluated on downstream tasks (often with a lightweight linear model on frozen features) ViT22B demonstrates increasing performance with scale. We further observe other interesting benefits of scale, including an improved tradeoff between bias and performance, an improved alignment to human visual perception in terms of shape/texture bias, and improved robustness. ViT22B demonstrates the potential for "LLM-like'' scaling in vision, and provides key steps towards getting there. View details
    Preview abstract Accurate uncertainty quantification is a major challenge in deep learning, as neural networks can make overconfident errors and assign high confidence predictions to out-of-distribution (OOD) inputs. The most popular approaches to estimate predictive uncertainty in deep learning are methods that combine predictions from multiple neural networks, such as Bayesian neural networks (BNNs) and deep ensembles. However their practicality in real-time, industrial-scale applications are limited due to the high memory and computational cost. Furthermore, ensembles and BNNs do not necessarily fix all the issues with the underlying member networks. In this work, we study principled approaches to improve uncertainty property of a single network, based on a single, deterministic representation. By formalizing the uncertainty quantification as a minimax learning problem, we first identify distance awareness, i.e., the model's ability to quantify the distance of a testing example from the training data, as a necessary condition for a DNN to achieve high-quality (i.e., minimax optimal) uncertainty estimation. We then propose Spectral-normalized Neural Gaussian Process (SNGP), a simple method that improves the distance-awareness ability of modern DNNs with two simple changes: (1) applying spectral normalization to hidden weights to enforce bi-Lipschitz smoothness in representations and (2) replacing the last output layer with a Gaussian process layer. On a suite of vision and language understanding benchmarks, SNGP outperforms other single-model approaches in prediction, calibration and out-of-domain detection. Furthermore, SNGP provides complementary benefits to popular techniques such as deep ensembles and data augmentation, making it a simple and scalable building block for probabilistic deep learning. Code is open-sourced at https://github.com/google/uncertainty-baselines. View details
    Plex: Towards Reliability using Pretrained Large Model Extensions
    Du Phan
    Mark Patrick Collier
    Zi Wang
    Zelda Mariet
    Clara Huiyi Hu
    Neil Band
    Tim G. J. Rudner
    Joost van Amersfoort
    Andreas Christian Kirsch
    Rodolphe Jenatton
    Honglin Yuan
    Kelly Buchanan
    Yarin Gal
    ICML 2022 Pre-training Workshop (2022)
    Preview abstract A recent trend in artificial intelligence (AI) is the use of pretrained models for language and vision tasks, which has achieved extraordinary performance but also puzzling failures. Examining tasks that probe the model’s abilities in diverse ways is therefore critical to the field. In this paper, we explore the \emph{reliability} of models, where we define a reliable model as one that not only achieves strong predictive performance but also performs well consistently over many decision-making tasks such as uncertainty (e.g., selective prediction, open set recognition), robust generalization (e.g., accuracy and scoring rules such as log-likelihood on in- and out-of-distribution datasets), and adaptation (e.g., active learning, few-shot learning). We devise 11 types of tasks over 36 datasets in order to evaluate different aspects of reliability on both vision and language domains. To improve reliability, we developed ViT-Plex and T5-Plex, \emph{p}retrained \emph{l}arge-model \emph{ex}tensions (henceforth abbreviated as \emph{plex}) for vision and language modalities. Plex greatly improves the state-of-the-art across tasks, and as a pretrained model Plex unifies the traditional protocol of designing and tuning one model for each reliability task. We demonstrate scaling effects over model sizes and pretraining dataset sizes up to 4 billion examples. We also demonstrate Plex’s capabilities on new tasks including zero-shot open set recognition, few-shot uncertainty, and uncertainty in conversational language understanding. View details
    Soft Calibration Objectives for Neural Networks
    Archit Karandikar
    Nick Cain
    Jon Shlens
    Michael C. Mozer
    Becca Roelofs
    Advances in Neural Information Processing Systems (NeurIPS) (2021)
    Preview abstract Optimal decision making requires that classifiers produce uncertainty estimates consistent with their empirical accuracy. However, deep neural networks are often under- or over-confident in their predictions. Consequently, methods have been developed to improve the calibration of their predictive uncertainty, both during training and post-hoc. In this work, we propose differentiable losses to improve calibration based on a soft (continuous) version of the binning operation underlying popular calibration-error estimators. When incorporated into training, these soft calibration losses achieve state-of-the-art single-model ECE across multiple datasets with less than 1% decrease in accuracy. For instance, we observe an 82% reduction in ECE (70% relative to the post-hoc rescaled ECE) in exchange for a 0.7% relative decrease in accuracy relative to the cross-entropy baseline on CIFAR-100. When incorporated post-training, the soft-binning-based calibration error objective improves upon temperature scaling, a popular recalibration method. Overall, experiments across losses and datasets demonstrate that using calibration- sensitive procedures yield better uncertainty estimates under dataset shift than the standard practice of using a cross-entropy loss and post-hoc recalibration methods. View details
    RecSim NG: Toward Principled Uncertainty Modeling for Recommender Ecosystems
    Martin Mladenov
    Vihan Jain
    Christopher Colby
    Nicolas Mayoraz
    Hubert Pham
    Ivan Vendrov
    ArXiv (2021)
    Preview abstract The development of recommender systems that optimize multi-turn interaction with users, and model the interactions of different agents (e.g., users, content providers, vendors) in the recommender ecosystem have drawn increasing attention in recent years. Developing and training models and algorithms for such recommenders can be especially difficult using static datasets, which often fail to offer the types of counterfactual predictions needed to evaluate policies over extended horizons. To address this, we develop RecSim NG, a probabilistic platform for the simulation of multi-agent recommender systems. RecSim NG is a scalable, modular, differentiable simulator implemented in Edward2 and TensorFlow. It offers: a powerful, general probabilistic programming language for agent-behavior specification; tools for probabilistic inference and latent-variable model learning, backed by automatic differentiation and tracing; a TensorFlow-based runtime for running simulations on accelerated hardware. We describe RecSim NG and illustrate how it can be used to create transparent, configurable, end-to-end models of a recommender ecosystem, complemented by a small set of simple use cases that demonstrate how RecSim NG can help both researchers and practitioners easily develop and train novel algorithms for recommender systems. A short version of this paper was published at RecSys 2020. View details
    Revisiting the Calibration of Modern Neural Networks
    Josip Djolonga
    Rob Romijnders
    Frances Ann Hubis
    Neural Information Processing Systems (2021) (to appear)
    Preview abstract Accurate estimation of predictive uncertainty (model calibration) is essential for the safe application of neural networks. Many instances of miscalibration in modern neural networks have been reported, suggesting a trend that newer, more accurate models produce poorly calibrated predictions. Here, we revisit this question for recent state-of-the-art image classification models. We systematically relate model calibration and accuracy, and find that the most recent models, notably those not using convolutions, are among the best calibrated. Trends observed in prior model generations, such as decay of calibration with distribution shift or model size, are less pronounced in recent architectures. We also show that model size and amount of pretraining do not fully explain these differences, suggesting that architecture is a major determinant of calibration properties. View details
    Deep Classifiers with Label Noise Modeling and Distance Awareness
    Vincent Fortuin
    Mark Patrick Collier
    Florian Wenzel
    James Urquhart Allingham
    Jesse Berent
    Rodolphe Jenatton
    NeurIPS 2021 Workshop on Bayesian Deep Learning (2021) (to appear)
    Preview abstract Uncertainty estimation in deep learning has recently emerged as a crucial area of interest to advance reliability and robustness of deep learning models, especially in safety-critical applications. While there have been many proposed methods that either focus on distance-aware model uncertainties for out-of-distribution detection or respectively on input-dependent label uncertainties for in-distribution calibration, combining these two approaches has been less well explored. In this work, we propose to combine these two ideas to achieve a joint modeling of model (epistemic) and data (aleatoric) uncertainty. We show that our combined model affords a favorable combination between these two complementary types of uncertainty and thus achieves good performance in-distribution and out-of-distribution on different benchmark datasets. View details
    Combining Ensembles and Data Augmentation Can Harm Your Calibration
    Yeming Wen
    Ghassen Jerfel
    Rafael Rios Müller
    International Conference on Learning Representations (2021)
    Preview abstract Ensemble methods which average over multiple neural network predictions are a simple approach to improve a model’s calibration and robustness. Similarly, data augmentation techniques, which encode prior information in the form of invariant feature transformations, are effective for improving calibration and robustness. In this paper, we show a surprising pathology: combining ensembles and data augmentation can harm model calibration. This leads to a trade-off in practice, whereby improved accuracy by combining the two techniques comes at the expense of calibration. On the other hand, selecting only one of the techniques ensures good uncertainty estimates at the expense of accuracy. We investigate this pathology and identify a compounding under-confidence among methods which marginalize over sets of weights and data augmentation techniques which soften labels. Finally, we propose a simple correction, achieving the best of both worlds with significant accuracy and calibration gains over using only ensembles or data augmentation individually. Applying the correction produces new state-of-the art in uncertainty calibration and robustness across CIFAR-10, CIFAR-100, and ImageNet. View details
    Training independent subnetworks for robust prediction
    Marton Havasi
    Rodolphe Jenatton
    Stanislav Fort
    International Conference on Learning Representations (2021)
    Preview abstract Recent approaches to efficiently ensemble neural networks have shown that strong robustness and uncertainty performance can be achieved with a negligible gain in parameters over the original network. However, these methods still require multiple forward passes for prediction, leading to a significant runtime cost. In this work, we show a surprising result: the benefits of using multiple predictions can be achieved 'for free' under a single model's forward pass. In particular, we show that, using a multi-input multi-output (MIMO) configuration, one can utilize a single model's capacity to train multiple subnetworks that independently learn the task at hand. By ensembling the predictions made by the subnetworks, we improve model robustness without increasing compute. We observe a significant improvement in negative log-likelihood, accuracy, and calibration error on CIFAR10, CIFAR100, ImageNet, and their out-of-distribution variants compared to previous methods. View details
    Hyperparameter Ensembles for Robustness and Uncertainty Quantification
    Florian Wenzel
    Rodolphe Jenatton
    Neural Information Processing Systems (NeurIPS) (2020)
    Preview abstract Ensembles over neural network weights trained from different random initialization, known as deep ensembles, achieve state-of-the-art accuracy and calibration. The recently introduced batch ensembles provide a drop-in replacement that is more parameter efficient. In this paper, we design ensembles not only over weights, but over hyperparameters to improve the state of the art in both settings. For best performance independent of budget, we propose hyper-deep ensembles, a simple procedure that involves a random search over different hyperparameters, themselves stratified across multiple random initializations. Its strong performance highlights the benefit of combining models with both weight and hyperparameter diversity. We further propose a parameter efficient version, hyper-batch ensembles, which builds on the layer structure of batch ensembles and self-tuning networks. The computational and memory costs of our method are notably lower than typical ensembles. On image classification tasks, with MLP, LeNet, ResNet 20 and Wide ResNet 28-10 architectures, we improve upon both deep and batch ensembles. View details
    Analyzing the Role of Model Uncertainty for Electronic Health Records
    Edward Choi
    Jeremy Nixon
    Ghassen Jerfel
    ACM Conference on Health, Inference, and Learning (ACM CHIL) (2020)
    Preview abstract In medicine, both ethical and monetary costs of incorrect predictions can be significant, and the complexity of the problems often necessitates increasingly complex models. Recent work has shown that changing just the random seed is enough for otherwise well-tuned deep neural networks to vary in their individual predicted probabilities. In light of this, we investigate the role of model uncertainty methods in the medical domain. Using RNN ensembles and various Bayesian RNNs, we show that population-level metrics, such as AUC-PR, AUC-ROC, log-likelihood, and calibration error, do not capture model uncertainty. Meanwhile, the presence of significant variability in patient-specific predictions and optimal decisions motivates the need for capturing model uncertainty. Understanding the uncertainty for individual patients is an area with clear clinical impact, such as determining when a model decision is likely to be brittle. We further show that RNNs with only Bayesian embeddings can be a more efficient way to capture model uncertainty compared to ensembles, and we analyze how model uncertainty is impacted across individual input features and patient subgroups. View details
    Demonstrating Principled Uncertainty Modeling for Recommender Ecosystems with RecSim NG
    Martin Mladenov
    Vihan Jain
    Christopher Colby
    Nicolas Mayoraz
    Hubert Pham
    Ivan Vendrov
    RecSys '20: Fourteenth ACM Conference on Recommender Systems (2020), pp. 591-593
    Preview abstract We develop RecSim NG, a probabilistic platform that supports natural, concise specification and learning of models for multi-agent recommender systems simulation. RecSim NG is a scalable, modular, differentiable simulator implemented in Edward2 and TensorFlow. An extended version of this paper is available as arXiv:2103.08057. View details
    BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong Learning
    Yeming Wen
    Jimmy Ba
    International Conference on Learning Representations (2020)
    Preview abstract Ensembles, where multiple neural networks are trained individually and their predictions are averaged, have been shown to be widely successful for improving both the accuracy and predictive uncertainty of single neural networks. However, an ensemble’s cost for both training and testing increases linearly with the number of networks, which quickly becomes untenable. In this paper, we propose BatchEnsemble, an ensemble method whose computational and memory costs are significantly lower than typical ensembles. BatchEnsemble achieves this by defining each weight matrix to be the Hadamard product of a shared weight among all ensemble members and a rank-one matrix per member. Unlike ensembles, BatchEnsemble is not only parallelizable across devices, where one device trains one member, but also parallelizable within a device, where multiple ensemble members are updated simultaneously for a given mini-batch. Across CIFAR-10, CIFAR-100, WMT14 EN-DE/EN-FR translation, and out-of-distribution tasks, BatchEnsemble yields competitive accuracy and uncertainties as typical ensembles; the speedup at test time is 3X and memory reduction is 3X at an ensemble of size 4. We also apply BatchEnsemble to lifelong learning, where on Split-CIFAR-100, BatchEnsemble yields comparable performance to progressive neural networks while having a much lower computational and memory costs. We further show that BatchEnsemble can easily scale up to lifelong learning on Split-ImageNet which involves 100 sequential learning tasks. View details
    Preview abstract Bayesian neural networks (BNN) and Deep Ensembles are principled approaches to estimate the predictive uncertainty of a deep learning model. However their practicality in real-time, industrial-scale applications are limited due to their heavy memory and inference cost. This motivates us to study principled approaches to high-quality uncertainty estimation that require only a single deep neural network (DNN). By formalizing the uncertainty quantification as a minimax learning problem, we first identify \textit{input distance awareness}, i.e., the model’s ability in quantifying the distance of a testing example from the training data in the input space, as a necessary condition for a DNN to achieve high-quality (i.e., minimax optimal) uncertainty estimation. We then propose \textit{Spectral-normalized Gaussian Process} (SNGP), a simple method that improves the distance-awareness ability of modern DNNs, by adding a weight normalization step during training and replacing the activation of the penultimate layer. We visually illustrate the property of the proposed method on two-dimensional datasets, and benchmark its performance against Deep Ensembles and other single-model approaches across both vision and language understanding tasks and on modern architectures (ResNet and BERT). Despite its simplicity, SNGP is competitive with Deep Ensembles in prediction, calibration and out-of-domain detection, and significantly outperforms the other single-model approaches. View details
    Preview abstract Bayesian neural networks (BNNs) demonstrate promising success in improving the robustness and uncertainty quantification of modern neural networks. However, they generally struggle with underfitting at scale and parameter efficiency. On the other hand, deep ensembles have emerged as an alternative for uncertainty quantification that, while outperforming BNNs on certain problems, also suffers from efficiency issues. It remains unclear how to combine the strengths of these two approaches and remediate their common issues. To tackle this challenge, we propose a rank-1 parameterization of BNNs, where each weight matrix involves only a distribution on a rank-1 subspace. We also revisit the use of mixture approximate posteriors to capture multiple modes where unlike typical mixtures, this approach admits a significantly smaller memory increase (e.g., only a 0.4% increase for a ResNet-50 mixture of size 10). We perform a systematic empirical study on the choices of prior, variational posterior, and methods to improve training. For ResNet-50 on ImageNet and Wide ResNet 28-10 on CIFAR-10/100, rank-1 BNNs outperform baselines across log-likelihood, accuracy, and calibration on the test set and out-of-distribution variants. View details
    Bayesian Layers: A Module for Neural Network Uncertainty
    Mark van der Wilk
    Danijar Hafner
    Neural Information Processing Systems (NeurIPS) (2019)
    Preview abstract We describe Bayesian Layers, a module designed for fast experimentation with neural network uncertainty. It extends neural network libraries with drop-in replacements for common layers. This enables composition via a unified abstraction over deterministic and stochastic functions and allows for scalability via the underlying system. These layers capture uncertainty over weights (Bayesian neural nets), pre-activation units (dropout), activations ("stochastic output layers"), or the function itself (Gaussian processes). They can also be reversible to propagate uncertainty from input to output. We include code examples for common architectures such as Bayesian LSTMs, deep GPs, and flow-based models. As demonstration, we fit a 5-billion parameter "Bayesian Transformer" on 512 TPUv2 cores for uncertainty in machine translation and a Bayesian dynamics model for model-based planning. Finally, we show how Bayesian Layers can be used within the Edward2 probabilistic programming language for probabilistic programs with stochastic processes. View details
    Discrete Flows: Invertible Generative Models of Discrete Data
    Keyon Vafa
    Kumar Krishna Agrawal
    Laurent Dinh
    NeurIPS (2019)
    Preview abstract While normalizing flows have led to significant advances in modeling high-dimensional continuous distributions, their applicability to discrete distributions remains unknown. In this paper, we show that flows can in fact be extended to discrete events---and under a simple change-of-variables formula not requiring log-determinant-Jacobian computations. Discrete flows have numerous applications. We consider two flow architectures: discrete autoregressive flows that enable bidirectionality, allowing, for example, tokens in text to depend on both left-to-right and right-to-left contexts in an exact language model; and discrete bipartite flows that enable efficient non-autoregressive generation as in RealNVP. Empirically, we find that discrete autoregressive flows outperform autoregressive baselines on synthetic discrete distributions, an addition task, and Potts models; and bipartite flows can obtain competitive performance with autoregressive baselines on character-level language modeling for Penn Tree Bank and text8. View details
    Preview abstract Hamiltonian Monte Carlo is a powerful algorithm for sampling from difficult-to-normalize posterior distributions. However, when the geometry of the posterior is unfavorable, it may take many expensive evaluations of the target distribution and its gradient to converge and mix. We propose neural transport (NeuTra) HMC, a technique for learning to correct this sort of unfavorable geometry using inverse autoregressive flows (IAF), a powerful neural variational inference technique. The IAF is trained to minimize the KL divergence from an isotropic Gaussian to the warped posterior, and then HMC sampling is performed in the warped space. We evaluate NeuTra HMC on a variety of synthetic and real problems, and find that it significantly outperforms vanilla HMC both in time to reach the stationary distribution and asymptotic effective-sample-size rates. View details
    Noise Contrastive Priors for Functional Uncertainty
    Danijar Hafner
    Timothy Lillicrap
    James Davidson
    UAI (2019)
    Preview abstract Obtaining reliable uncertainty estimates of neural network predictions is a long standing challenge. Bayesian neural networks have been proposed as a solution, but it remains open how to specify their prior. In particular, the common practice of an independent normal prior in weight space imposes relatively weak constraints on the function posterior, allowing it to generalize in unforeseen ways on inputs outside of the training distribution. We propose noise contrastive priors (NCPs) to obtain reliable uncertainty estimates. The key idea is to train the model to output high uncertainty for data points outside of the training distribution. NCPs do so using an input prior, which adds noise to the inputs of the current mini batch, and an output prior, which is a wide distribution given these inputs. NCPs are compatible with any model that can output uncertainty estimates, are easy to scale, and yield reliable uncertainty estimates throughout training. Empirically, we show that NCPs prevent overfitting outside of the training distribution and result in uncertainty estimates that are useful for active learning. We demonstrate the scalability of our method on the flight delays data set, where we significantly improve upon previously published results. View details
    Preview abstract Deriving conditional and marginal distributions using conjugacy relationships can be time consuming and error prone. In this paper, we propose a strategy for automating such derivations. Unlike previous systems which focus on relationships between pairs of random variables, our system (which we call Autoconj) operates directly on Python functions that compute log-joint distribution functions. Autoconj provides support for conjugacy-exploiting algorithms in any Python-embedded PPL. This paves the way for accelerating development of novel inference algorithms and structure-exploiting modeling strategies. The package can be downloaded at https://github.com/google-research/autoconj. View details
    Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches
    Yeming Wen
    Paul Vicol
    Jimmy Ba
    Roger Grosse
    International Conference on Learning Representations (2018)
    Preview abstract Stochastic neural net weights are used in a variety of contexts, including regularization, Bayesian neural nets, exploration in reinforcement learning, and evolution strategies. Unfortunately, due to the large number of weights, all the examples in a mini-batch typically share the same weight perturbation, thereby limiting the variance reduction effect of large mini-batches. We introduce flipout, an efficient method for decorrelating the gradients within a mini-batch by implicitly sampling pseudo-independent weight perturbations for each example. Empirically, flipout achieves the ideal linear variance reduction for fully connected networks, convolutional networks, and RNNs. We find significant speedups in training neural networks with multiplicative Gaussian perturbations. We show that flipout is effective at regularizing LSTMs, and outperforms previous methods. Flipout also enables us to vectorize evolution strategies: in our experiments, a single GPU with flipout can handle the same throughput as at least 40 CPU cores using existing methods, equivalent to a factor-of-4 cost reduction on Amazon Web Services. View details
    Preview abstract Deriving conditional and marginal distributions using conjugacy relationships can be time consuming and error prone. In this paper, we propose a strategy for automating such derivations. Unlike previous systems which focus on relationships between pairs of random variables, our system (which we call Autoconj) operates directly on Python functions that compute log-joint distribution functions. Autoconj provides support for conjugacy-exploiting algorithms in any Python-embedded PPL. This paves the way for accelerating development of novel inference algorithms and structure-exploiting modeling strategies. The package can be downloaded at https://github.com/google-research/autoconj. View details
    Preview abstract We describe Edward2, a low-level probabilistic programming language. Edward2 distills the core of probabilistic programming down to a single abstraction—the random variable. By blurring the line between model and computation, Edward2 enables numerous applications not shown before: a model-parallel variational auto-encoder (VAE) with tensor processing units (TPUs); a data-parallel autoregressive model (Image Transformer) with TPUs; and multi-GPU No-U-Turn Sampler (NUTS). Edward2 achieves an optimal linear speedup from 4 to 256 TPUs. With VAEs, Edward2 sees up to a 20x speedup on TPUs over Pyro and Edward on GPUs; with Bayesian neural networks, Edward2 sees up to a 51x speedup. With NUTS, Edward2 sees a 20x speedup on GPUs over Stan and 7x over PyMC3. View details
    Image Transformer
    Niki J. Parmar
    Ashish Vaswani
    Jakob Uszkoreit
    Lukasz Kaiser
    Noam Shazeer
    International Conference on Machine Learning (ICML) (2018)
    Preview abstract Recent work demonstrated significant progress towards modeling the distribution of natural images with tractable likelihood using deep neural networks. This was achieved by modeling the joint distribution of pixels in the image as the product of conditional distributions, thereby turning it into a sequence modeling problem, and applying recurrent or convolutional neural networks to it. In this work we instead build on the Transformer, a recently proposed network architecture based on self-attention, to model the conditional distributions in similar factorizations. We present two extensions of the network architecture, allowing it to scale to images and to take advantage of their two-dimensional structure. While conceptually simple, our generative models trained on two image data sets are competitive with or outperform the current state of the art on two different data sets, CIFAR-10 and ImageNet, as measured by log-likelihood. We also present results on image super-resolution with large magnification ratio with an encoder-decoder configuration of our architecture. In a human evaluation study, we show that our super-resolution models improve over previously published autoregressive super-resolution models in how often they fool a naive human observer by a factor of three. Lastly, we provide examples of images generated or completed by our various models which, following previous work, we also believe to look pretty cool. View details
    Mesh-TensorFlow: Deep Learning for Supercomputers
    Noam Shazeer
    Youlong Cheng
    Niki J. Parmar
    Ashish Vaswani
    Peter Hawkins
    HyoukJoong Lee
    Mingsheng Hong
    Ryan Sepassi
    Black Hechtman
    NeurIPS (2018)
    Preview abstract Batch-splitting (data-parallelism) is the dominant distributed Deep Neural Network (DNN) training strategy, due to its universal applicability and its amenability to Single-Program-Multiple-Data (SPMD) programming. However, batch-splitting suffers from problems including the inability to train very large models (due to memory constraints), high latency, and inefficiency at small batch sizes. All of these can be solved by more general distribution strategies (model-parallelism). Unfortunately, efficient model-parallel algorithms tend to be complicated to discover, describe, and to implement, particularly on large clusters. We introduce Mesh-TensorFlow, a language for specifying a general class of distributed tensor computations. Where data-parallelism can be viewed as splitting tensors and operations along the "batch" dimension, in Mesh-TensorFlow, the user can specify any tensor-dimensions to be split across any dimensions of a multi-dimensional mesh of processors. A Mesh-TensorFlow graph compiles into a SPMD program consisting of parallel operations coupled with collective communication primitives such as Allreduce. We use Mesh-TensorFlow to implement an efficient data-parallel, model-parallel version of the Transformer sequence-to-sequence model. Using TPU meshes of up to 512 cores, we train Transformer models with up to 5 billion parameters, surpassing SOTA results on WMT'14 English-to-French translation task and the one-billion-word Language modeling benchmark. Mesh-Tensorflow is available at https://github.com/tensorflow/mesh View details
    No Results Found