Searcharxiv⌕ Search

arXiv subjects

Jerry Li

Publications and source records attributed to Jerry Li.

At least 109 records · Page 6Linked to original sources

On Mean Estimation for General Norms with Statistical Queries

We study the problem of mean estimation for high-dimensional distributions, assuming access to a statistical query oracle for the distribution. For a normed space $X = (\mathbb{R}^d, \|\cdot\|_X)$ and a distribution supported on vectors $x \in \mathbb{R}^d$ with $\|x\|_{X} \leq 1$, the task is to output an estimate $\hatμ \in \mathbb{R}^d$ which is $ε$-close in the distance induced by $\|\cdot\|_X$ to the true mean of the distribution. We obtain sharp upper and lower bounds for the statistical query complexity of this problem when the the underlying norm is symmetric as well as for Schatten-$p$ norms, answering two questions raised by Feldman, Guzmán, and Vempala (SODA 2017).

cs.DS↗

Spectral Signatures in Backdoor Attacks

A recent line of work has uncovered a new form of data poisoning: so-called \emph{backdoor} attacks. These attacks are particularly dangerous because they do not affect a network's behavior on typical, benign data. Rather, the network only deviates from its expected output when triggered by a perturbation planted by an adversary. In this paper, we identify a new property of all known backdoor attacks, which we call \emph{spectral signatures}. This property allows us to utilize tools from robust statistics to thwart the attacks. We demonstrate the efficacy of these signatures in detecting and removing poisoned examples on real image sets and state of the art neural network architectures. We believe that understanding spectral signatures is a crucial first step towards designing ML systems secure against such backdoor attacks

cs.LG↗

Twin-GAN -- Unpaired Cross-Domain Image Translation with Weight-Sharing GANs

We present a framework for translating unlabeled images from one domain into analog images in another domain. We employ a progressively growing skip-connected encoder-generator structure and train it with a GAN loss for realistic output, a cycle consistency loss for maintaining same-domain translation identity, and a semantic consistency loss that encourages the network to keep the input semantic features in the output. We apply our framework on the task of translating face images, and show that it is capable of learning semantic mappings for face images with no supervised one-to-one image mapping.

cs.CV↗

On the Limitations of First-Order Approximation in GAN Dynamics

While Generative Adversarial Networks (GANs) have demonstrated promising performance on multiple vision tasks, their learning dynamics are not yet well understood, both in theory and in practice. To address this issue, we study GAN dynamics in a simple yet rich parametric model that exhibits several of the common problematic convergence behaviors such as vanishing gradients, mode collapse, and diverging or oscillatory behavior. In spite of the non-convex nature of our model, we are able to perform a rigorous theoretical analysis of its convergence behavior. Our analysis reveals an interesting dichotomy: a GAN with an optimal discriminator provably converges, while first order approximations of the discriminator steps lead to unstable GAN dynamics and mode collapse. Our result suggests that using first order discriminator steps (the de-facto standard in most existing GAN setups) might be one of the factors that makes GAN training challenging in practice.

cs.LG↗

Byzantine Stochastic Gradient Descent

This paper studies the problem of distributed stochastic optimization in an adversarial setting where, out of the $m$ machines which allegedly compute stochastic gradients every iteration, an $α$-fraction are Byzantine, and can behave arbitrarily and adversarially. Our main result is a variant of stochastic gradient descent (SGD) which finds $\varepsilon$-approximate minimizers of convex functions in $T = \tilde{O}\big( \frac{1}{\varepsilon^2 m} + \frac{α^2}{\varepsilon^2} \big)$ iterations. In contrast, traditional mini-batch SGD needs $T = O\big( \frac{1}{\varepsilon^2 m} \big)$ iterations, but cannot tolerate Byzantine failures. Further, we provide a lower bound showing that, up to logarithmic factors, our algorithm is information-theoretically optimal both in terms of sampling complexity and time complexity.

cs.LG↗

Being Robust (in High Dimensions) Can Be Practical

Robust estimation is much more challenging in high dimensions than it is in one dimension: Most techniques either lead to intractable optimization problems or estimators that can tolerate only a tiny fraction of errors. Recent work in theoretical computer science has shown that, in appropriate distributional models, it is possible to robustly estimate the mean and covariance with polynomial time algorithms that can tolerate a constant fraction of corruptions, independent of the dimension. However, the sample and time complexity of these algorithms is prohibitively large for high-dimensional applications. In this work, we address both of these issues by establishing sample complexity bounds that are optimal, up to logarithmic factors, as well as giving various refinements that allow the algorithms to tolerate a much larger fraction of corruptions. Finally, we show on both synthetic and real data that our algorithms have state-of-the-art performance and suddenly make high-dimensional robust estimation a realistic possibility.

cs.LG↗

Fast and Sample Near-Optimal Algorithms for Learning Multidimensional Histograms

We study the problem of robustly learning multi-dimensional histograms. A $d$-dimensional function $h: D \rightarrow \mathbb{R}$ is called a $k$-histogram if there exists a partition of the domain $D \subseteq \mathbb{R}^d$ into $k$ axis-aligned rectangles such that $h$ is constant within each such rectangle. Let $f: D \rightarrow \mathbb{R}$ be a $d$-dimensional probability density function and suppose that $f$ is $\mathrm{OPT}$-close, in $L_1$-distance, to an unknown $k$-histogram (with unknown partition). Our goal is to output a hypothesis that is $O(\mathrm{OPT}) + ε$ close to $f$, in $L_1$-distance. We give an algorithm for this learning problem that uses $n = \tilde{O}_d(k/ε^2)$ samples and runs in time $\tilde{O}_d(n)$. For any fixed dimension, our algorithm has optimal sample complexity, up to logarithmic factors, and runs in near-linear time. Prior to our work, the time complexity of the $d=1$ case was well-understood, but significant gaps in our understanding remained even for $d=2$.

cs.LG↗

QSGD: Communication-Efficient SGD via Gradient Quantization and Encoding

Parallel implementations of stochastic gradient descent (SGD) have received significant research attention, thanks to excellent scalability properties of this algorithm, and to its efficiency in the context of training deep neural networks. A fundamental barrier for parallelizing large-scale SGD is the fact that the cost of communicating the gradient updates between nodes can be very large. Consequently, lossy compression heuristics have been proposed, by which nodes only communicate quantized gradients. Although effective in practice, these heuristics do not always provably converge, and it is not clear whether they are optimal. In this paper, we propose Quantized SGD (QSGD), a family of compression schemes which allow the compression of gradient updates at each node, while guaranteeing convergence under standard assumptions. QSGD allows the user to trade off compression and convergence time: it can communicate a sublinear number of bits per iteration in the model dimension, and can achieve asymptotically optimal communication cost. We complement our theoretical results with empirical data, showing that QSGD can significantly reduce communication cost, while being competitive with standard uncompressed techniques on a variety of real tasks. In particular, experiments show that gradient quantization applied to training of deep neural networks for image classification and automated speech recognition can lead to significant reductions in communication cost, and end-to-end training time. For instance, on 16 GPUs, we are able to train a ResNet-152 network on ImageNet 1.8x faster to full accuracy. Of note, we show that there exist generic parameter settings under which all known network architectures preserve or slightly improve their full accuracy when using quantization.

cs.LG↗

Mixture Models, Robustness, and Sum of Squares Proofs

We use the Sum of Squares method to develop new efficient algorithms for learning well-separated mixtures of Gaussians and robust mean estimation, both in high dimensions, that substantially improve upon the statistical guarantees achieved by previous efficient algorithms. Firstly, we study mixtures of $k$ distributions in $d$ dimensions, where the means of every pair of distributions are separated by at least $k^{\varepsilon}$. In the special case of spherical Gaussian mixtures, we give a $(dk)^{O(1/\varepsilon^2)}$-time algorithm that learns the means assuming separation at least $k^{\varepsilon}$, for any $\varepsilon > 0$. This is the first algorithm to improve on greedy ("single-linkage") and spectral clustering, breaking a long-standing barrier for efficient algorithms at separation $k^{1/4}$. We also study robust estimation. When an unknown $(1-\varepsilon)$-fraction of $X_1,\ldots,X_n$ are chosen from a sub-Gaussian distribution with mean $μ$ but the remaining points are chosen adversarially, we give an algorithm recovering $μ$ to error $\varepsilon^{1-1/t}$ in time $d^{O(t^2)}$, so long as sub-Gaussian-ness up to $O(t)$ moments can be certified by a Sum of Squares proof. This is the first polynomial-time algorithm with guarantees approaching the information-theoretic limit for non-Gaussian distributions. Previous algorithms could not achieve error better than $\varepsilon^{1/2}$. Both of these results are based on a unified technique. Inspired by recent algorithms of Diakonikolas et al. in robust statistics, we devise an SDP based on the Sum of Squares method for the following setting: given $X_1,\ldots,X_n \in \mathbb{R}^d$ for large $d$ and $n = poly(d)$ with the promise that a subset of $X_1,\ldots,X_n$ were sampled from a probability distribution with bounded moments, recover some information about that distribution.

cs.DS↗

Robustly Learning a Gaussian: Getting Optimal Error, Efficiently

We study the fundamental problem of learning the parameters of a high-dimensional Gaussian in the presence of noise -- where an $\varepsilon$-fraction of our samples were chosen by an adversary. We give robust estimators that achieve estimation error $O(\varepsilon)$ in the total variation distance, which is optimal up to a universal constant that is independent of the dimension. In the case where just the mean is unknown, our robustness guarantee is optimal up to a factor of $\sqrt{2}$ and the running time is polynomial in $d$ and $1/ε$. When both the mean and covariance are unknown, the running time is polynomial in $d$ and quasipolynomial in $1/\varepsilon$. Moreover all of our algorithms require only a polynomial number of samples. Our work shows that the same sorts of error guarantees that were established over fifty years ago in the one-dimensional setting can also be achieved by efficient algorithms in high-dimensional settings.

cs.DS↗

The ZipML Framework for Training Models with End-to-End Low Precision: The Cans, the Cannots, and a Little Bit of Deep Learning

Recently there has been significant interest in training machine-learning models at low precision: by reducing precision, one can reduce computation and communication by one order of magnitude. We examine training at reduced precision, both from a theoretical and practical perspective, and ask: is it possible to train models at end-to-end low precision with provable guarantees? Can this lead to consistent order-of-magnitude speedups? We present a framework called ZipML to answer these questions. For linear models, the answer is yes. We develop a simple framework based on one simple but novel strategy called double sampling. Our framework is able to execute training at low precision with no bias, guaranteeing convergence, whereas naive quantization would introduce significant bias. We validate our framework across a range of applications, and show that it enables an FPGA prototype that is up to 6.5x faster than an implementation using full 32-bit precision. We further develop a variance-optimal stochastic quantization strategy and show that it can make a significant difference in a variety of settings. When applied to linear models together with double sampling, we save up to another 1.7x in data movement compared with uniform quantization. When training deep networks with quantized models, we achieve higher accuracy than the state-of-the-art XNOR-Net. Finally, we extend our framework through approximation to non-linear models, such as SVM. We show that, although using low-precision data induces bias, we can appropriately bound and control the bias. We find in practice 8-bit precision is often sufficient to converge to the correct solution. Interestingly, however, in practice we notice that our framework does not always outperform the naive rounding approach. We discuss this negative result in detail.

cs.LG↗

The Power of Choice in Priority Scheduling

Consider the following random process: we are given $n$ queues, into which elements of increasing labels are inserted uniformly at random. To remove an element, we pick two queues at random, and remove the element of lower label (higher priority) among the two. The cost of a removal is the rank of the label removed, among labels still present in any of the queues, that is, the distance from the optimal choice at each step. Variants of this strategy are prevalent in state-of-the-art concurrent priority queue implementations. Nonetheless, it is not known whether such implementations provide any rank guarantees, even in a sequential model. We answer this question, showing that this strategy provides surprisingly strong guarantees: Although the single-choice process, where we always insert and remove from a single randomly chosen queue, has degrading cost, going to infinity as we increase the number of steps, in the two choice process, the expected rank of a removed element is $O( n )$ while the expected worst-case cost is $O( n \log n )$. These bounds are tight, and hold irrespective of the number of steps for which we run the process. The argument is based on a new technical connection between "heavily loaded" balls-into-bins processes and priority scheduling. Our analytic results inspire a new concurrent priority queue implementation, which improves upon the state of the art in terms of practical performance.

cs.DS↗

Robust Sparse Estimation Tasks in High Dimensions

In this paper we initiate the study of whether or not sparse estimation tasks can be performed efficiently in high dimensions, in the robust setting where an $\eps$-fraction of samples are corrupted adversarially. We study the natural robust version of two classical sparse estimation problems, namely, sparse mean estimation and sparse PCA in the spiked covariance model. For both of these problems, we provide the first efficient algorithms that provide non-trivial error guarantees in the presence of noise, using only a number of samples which is similar to the number required for these problems without noise. In particular, our sample complexities are sublinear in the ambient dimension $d$. Our work also suggests evidence for new computational-vs-statistical gaps for these problems (similar to those for sparse PCA without noise) which only arise in the presence of noise.

cs.LG↗

Fast Algorithms for Segmented Regression

We study the fixed design segmented regression problem: Given noisy samples from a piecewise linear function $f$, we want to recover $f$ up to a desired accuracy in mean-squared error. Previous rigorous approaches for this problem rely on dynamic programming (DP) and, while sample efficient, have running time quadratic in the sample size. As our main contribution, we provide new sample near-linear time algorithms for the problem that -- while not being minimax optimal -- achieve a significantly better sample-time tradeoff on large datasets compared to the DP approach. Our experimental evaluation shows that, compared with the DP approach, our algorithms provide a convergence rate that is only off by a factor of $2$ to $4$, while achieving speedups of three orders of magnitude.

cs.LG↗

A Nearly Optimal and Agnostic Algorithm for Properly Learning a Mixture of k Gaussians, for any Constant k

Learning a Gaussian mixture model (GMM) is a fundamental problem in machine learning, learning theory, and statistics. One notion of learning a GMM is proper learning: here, the goal is to find a mixture of $k$ Gaussians $\mathcal{M}$ that is close to the density $f$ of the unknown distribution from which we draw samples. The distance between $\mathcal{M}$ and $f$ is typically measured in the total variation or $L_1$-norm. We give an algorithm for learning a mixture of $k$ univariate Gaussians that is nearly optimal for any fixed $k$. The sample complexity of our algorithm is $\tilde{O}(\frac{k}{ε^2})$ and the running time is $(k \cdot \log\frac{1}ε)^{O(k^4)} + \tilde{O}(\frac{k}{ε^2})$. It is well-known that this sample complexity is optimal (up to logarithmic factors), and it was already achieved by prior work. However, the best known time complexity for proper learning a $k$-GMM was $\tilde{O}(\frac{1}{ε^{3k-1}})$. In particular, the dependence between $\frac{1}ε$ and $k$ was exponential. We significantly improve this dependence by replacing the $\frac{1}ε$ term with a $\log \frac{1}ε$ while only increasing the exponent moderately. Hence, for any fixed $k$, the $\tilde{O} (\frac{k}{ε^2})$ term dominates our running time, and thus our algorithm runs in time which is nearly-linear in the number of samples drawn. Achieving a running time of $\textrm{poly}(k, \frac{1}ε)$ for proper learning of $k$-GMMs has recently been stated as an open problem by multiple researchers, and we make progress on this question. Moreover, our approach offers an agnostic learning guarantee: our algorithm returns a good GMM even if the distribution we are sampling from is not a mixture of Gaussians. To the best of our knowledge, our algorithm is the first agnostic proper learning algorithm for GMMs.

cs.DS↗

Sample-Optimal Density Estimation in Nearly-Linear Time

We design a new, fast algorithm for agnostically learning univariate probability distributions whose densities are well approximated by piecewise polynomial functions. Let $f$ be the density function of an arbitrary univariate distribution, and suppose that $f$ is $\mathrm{OPT}$-close in $L_1$-distance to an unknown piecewise polynomial function with $t$ interval pieces and degree $d$. Our algorithm draws $n = O(t(d+1)/ε^2)$ samples from $f$, runs in time $\tilde{O}(n \cdot \mathrm{poly}(d))$, and with probability at least $9/10$ outputs an $O(t)$-piecewise degree-$d$ hypothesis $h$ that is $4 \cdot \mathrm{OPT} +ε$ close to $f$. Our general algorithm yields (nearly) sample-optimal and nearly-linear time estimators for a wide range of structured distribution families over both continuous and discrete domains in a unified way. For most of our applications, these are the first sample-optimal and nearly-linear time estimators in the literature. As a consequence, our work resolves the sample and computational complexities of a broad class of inference tasks via a single "meta-algorithm". Moreover, we experimentally demonstrate that our algorithm performs very well in practice. Our algorithm consists of three "levels": (i) At the top level, we employ an iterative greedy algorithm for finding a good partition of the real line into the pieces of a piecewise polynomial. (ii) For each piece, we show that the sub-problem of finding a good polynomial fit on the current interval can be solved efficiently with a separation oracle method. (iii) We reduce the task of finding a separating hyperplane to a combinatorial problem and give an efficient algorithm for this problem. Combining these three procedures gives a density estimation algorithm with the claimed guarantees.

cs.DS↗

Replacing Mark Bits with Randomness in Fibonacci Heaps

A Fibonacci heap is a deterministic data structure implementing a priority queue with optimal amortized operation costs. An unfortunate aspect of Fibonacci heaps is that they must maintain a "mark bit" which serves only to ensure efficiency of heap operations, not correctness. Karger proposed a simple randomized variant of Fibonacci heaps in which mark bits are replaced by coin flips. This variant still has expected amortized cost $O(1)$ for insert, decrease-key, and merge. Karger conjectured that this data structure has expected amortized cost $O(\log s)$ for delete-min, where $s$ is the number of heap operations. We give a tight analysis of Karger's randomized Fibonacci heaps, resolving Karger's conjecture. Specifically, we obtain matching upper and lower bounds of $Θ(\log^2 s / \log \log s)$ for the runtime of delete-min. We also prove a tight lower bound of $Ω(\sqrt{n})$ on delete-min in terms of the number of heap elements $n$. The request sequence used to prove this bound also solves an open problem of Fredman on whether cascading cuts are necessary. Finally, we give a simple additional modification to these heaps which yields a tight runtime $O(\log^2 n / \log \log n)$ for delete-min.

cs.DS↗

On the Importance of Registers for Computability

All consensus hierarchies in the literature assume that we have, in addition to copies of a given object, an unbounded number of registers. But why do we really need these registers? This paper considers what would happen if one attempts to solve consensus using various objects but without any registers. We show that under a reasonable assumption, objects like queues and stacks cannot emulate the missing registers. We also show that, perhaps surprisingly, initialization, shown to have no computational consequences when registers are readily available, is crucial in determining the synchronization power of objects when no registers are allowed. Finally, we show that without registers, the number of available objects affects the level of consensus that can be solved. Our work thus raises the question of whether consensus hierarchies which assume an unbounded number of registers truly capture synchronization power, and begins a line of research aimed at better understanding the interaction between read-write memory and the powerful synchronization operations available on modern architectures.

cs.DC↗