Zachary Nado
I’m Zachary Nado, a Research Engineer at Google Brain in Cambridge, MA where our team works on anything and everything related to machine learning and artificial intelligence!
I graduated from Brown University in Computer Science and Applied Mathematics in 2016 where I was a part of the Serre Lab. There I worked on various systems engineering problems for the lab, lead a team of six to design a web annotation tool for labeling and viewing machine learning data, and developed my honors thesis to replace an older computer vision pipeline for classifying mouse behavior with convolutional neural networks.
During my college summers I did two internships with Google where I worked on several search infrastructure projects, followed by an internship at SpaceX as part of their software engineering team.
I was also a member of the Brown Space Engineering team for three years where I worked on our first ever satellite, a 1U cubesat that acts as an artificial shooting star that was launched to orbit in May 2018.
Authored Publications
Sort By
Adaptive Gradient Methods at the Edge of Stability
Behrooz Ghorbani
David Cardoze
Jeremy Cohen
Justin Gilmer
Shankar Krishnan
NeuRIPS 2022 (2022) (to appear)
Preview abstract
Little is known about the training dynamics of adaptive gradient methods like Adam in deep learning. In this paper, we shed light on the behavior of these algorithms in the full-batch and sufficiently large batch settings. Specifically, we show that during full-batch training, the maximum eigenvalue of the \emph{preconditioned} Hessian typically equilibrates at the stability threshold of a related non-adaptive algorithm. For Adam with step size $\eta$ and $\beta_1 = 0.9$, this stability threshold is $38/\eta$. Similar effects occur during minibatch training, especially as the batch size grows. Yet, even though adaptive methods train at the “Edge of Stability,” their behavior in this regime differs in a crucial way from that of their non-adaptive counterparts. Whereas non-adaptive algorithms are forced to remain in low-curvature regions of the loss landscape, we demonstrate that adaptive gradient methods often advance into high-curvature regions, while adapting the preconditioner to compensate. We believe that our findings will serve as a foundation for the community’s future understanding of adaptive gradient methods in deep learning.
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
Karan Singhal
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
A Loss Curvature Perspective On Training Instability in Deep Learning
Justin Gilmer
Behrooz Ghorbani
Ankush Garg
Behnam Neyshabur
David Cardoze
ICLR (2022)
Preview abstract
In this work, we study the evolution of the loss Hessian across many classification tasks in order to understand the effect the curvature of the loss has on the training dynamics. Whereas prior work has focused on how different learning rates affect the loss Hessian observed during training, we also analyze the effects of model initialization, architectural choices, and common training heuristics such as gradient clipping and learning rate warmup. Our results demonstrate that successful model and hyperparameter choices allow the early optimization trajectory to either avoid---or navigate out of---regions of high curvature and into flatter regions that tolerate a higher learning rate. Our results suggest a unifying perspective on how disparate mitigation strategies for training instability ultimately address the same underlying failure mode of neural network optimization, namely poor conditioning. Inspired by the conditioning perspective, we show that learning rate warmup can improve training stability just as much as batch normalization, layer normalization, MetaInit, GradInit, and Fixup initialization.
View details
A Simple Approach to Improve Single-Model Deep Uncertainty via Distance-Awareness
Shreyas Padhy
Zi Lin
Yeming Wen
Ghassen Jerfel
Journal of Machine Learning Research (2022)
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
Revisiting One-vs-All Classifiers for Predictive Uncertainty and Out-of-Distribution Detection in Neural Networks
Shreyas Padhy
(2020)
Preview abstract
Accurate estimation of predictive uncertainty in modern neural networks is critical to achieve well calibrated predictions and detect out-of-distribution inputs. The most promising approaches have been predominantly focused on improving model uncertainty (e.g. deep ensembles and Bayesian neural networks) and post-processing techniques for out-of-distribution detection (e.g. ODIN and Mahalanobis distance). However, there has been relatively little investigation into how the parametrization of the probabilities in discriminative classifiers affects the uncertainty estimates, and the dominant method, softmax cross-entropy, results in misleadingly high confidences on out-of-distribution data and under covariate shift. We investigate alternative ways of formulating probabilities using (1) a one-vs-all formulation to capture the notion of “none of the above”, and (2) a distance-based logit representation to encode uncertainty as a function of distance to the training manifold. We show that one-vs-all formulations can match the predictive performance of softmax without incurring any additional training or test-time complexity, and improve calibration on image classification tasks.
View details
Underspecification Presents Challenges for Credibility in Modern Machine Learning
Dan Moldovan
Ben Adlam
Babak Alipanahi
Alex Beutel
Christina Chen
Jon Deaton
Matthew D. Hoffman
Shaobo Hou
Neil Houlsby
Ghassen Jerfel
Yian Ma
Diana Mincu
Akinori Mitani
Andrea Montanari
Christopher Nielsen
Thomas Osborne
Rajiv Raman
Kim Ramasamy
Martin Gamunu Seneviratne
Shannon Sequeira
Harini Suresh
Victor Veitch
Steve Yadlowsky
Xiaohua Zhai
Journal of Machine Learning Research (2020)
Preview abstract
ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.
View details
Preview abstract
Traditionally there has been a perceived trade-off between machine learning code that is easy to write and machine learning code that fast, scalable, or easy to distribute, with platforms like TensorFlow, Theano, PyTorch, and Autograd inhabiting different points along this tradeoff curve. PyTorch and Autograd offer the coding benefits of imperative programming style and accept the computational tradeoff of interpretive overhead. TensorFlow and Theano give the benefit of whole-program optimization based on defined computation graphs, with the trade-off of potentially cumbersome graph-based semantics and associated developer overhead, which become especially apparent for more complex model types that depend on control flow operators. We propose to capture the benefits of both paradigms, using imperative programming style while enabling high performance program optimization, by using staged programming via source code transformation to essentially compile native Python into a lower-level IR like TensorFlow graphs. A key insight is to delay all type-dependent decisions until runtime, via dynamic dispatch. We instantiate these principles in AutoGraph, a piece of software that improves the programming experience of the TensorFlow machine learning library, and demonstrate the strong usability improvements with no loss in performance compared to native TensorFlow graphs.\end{abstract}
View details
Can You Trust Your Model’s Uncertainty? Evaluating Predictive Uncertainty Under Dataset Shift
Yaniv Ovadia
Sebastian Nowozin
Josh Dillon
Advances in Neural Information Processing Systems (2019)
Preview abstract
Modern machine learning methods including deep learning have achieved great success in predictive accuracy for supervised learning tasks, but may still fall short in giving useful estimates of their predictive {\em uncertainty}. Quantifying uncertainty is especially critical in real-world settings, which often involve distributions that are skewed from the training distribution due to a variety of factors including sample bias and non-stationarity. In such settings, well calibrated uncertainty estimates convey information about when a model's output should (or should not) be trusted. Many probabilistic deep learning methods, including Bayesian-and non-Bayesian methods, have been proposed in the literature for quantifying predictive uncertainty, but to our knowledge there has not previously been a rigorous large-scale empirical comparison of these methods under conditions of distributional skew. We present a large-scale benchmark of existing state-of-the-art methods on classification problems and investigate the effect of distributional skew on accuracy and calibration. We find that traditional post-hoc calibration falls short and some Bayesian methods are intractable for very large data. However, methods that marginalize over models give surprisingly strong results across a broad spectrum.
View details
Which Algorithmic Choices Matter at Which Batch Sizes? Insights From a Noisy Quadratic Model
Guodong Zhang
James Martens
Sushant Sachdeva
Chris Shallue
Roger Grosse
2019 Conference on Neural Information Processing Systems (2019)
Preview abstract
Increasing the batch size is a popular way to speed up neural network training, but beyond some critical batch size, larger batch sizes yield diminishing returns. In this work, we study how the critical batch size changes based on properties of the optimization algorithm, including acceleration and preconditioning, through two different lenses: large scale experiments, and analysis of a simple noisy quadratic model (NQM). We experimentally demonstrate that optimization algorithms that employ preconditioning, specifically Adam and K-FAC, result in much larger critical batch sizes than stochastic gradient descent with momentum. We also demonstrate that the NQM captures many of the essential features of real neural network training, despite being drastically simpler to work with. The NQM predicts our results with preconditioned optimizers, previous results with accelerated gradient descent, and other results around optimal learning rates and large batch training, making it a useful tool to generate testable predictions about neural network optimization.
View details
