• Binary embedding is a nonlinear dimension reduction methodology where high dimensional data are embedded into the Hamming cube while preserving the structure of the original space. Specifically, for an arbitrary $N$ distinct points in $\mathbb{S}^{p-1}$, our goal is to encode each point using $m$-dimensional binary strings such that we can reconstruct their geodesic distance up to $\delta$ uniform distortion. Existing binary embedding algorithms either lack theoretical guarantees or suffer from running time $O\big(mp\big)$. We make three contributions: (1) we establish a lower bound that shows any binary embedding oblivious to the set of points requires $m = \Omega(\frac{1}{\delta^2}\log{N})$ bits and a similar lower bound for non-oblivious embeddings into Hamming distance; (2) [DELETED, see comment]; (3) we also provide an analytic result about embedding a general set of points $K \subseteq \mathbb{S}^{p-1}$ with even infinite size. Our theoretical findings are supported through experiments on both synthetic and real data sets.
  • We study the problem of testing identity against a given distribution with a focus on the high confidence regime. More precisely, given samples from an unknown distribution $p$ over $n$ elements, an explicitly given distribution $q$, and parameters $0< \epsilon, \delta < 1$, we wish to distinguish, {\em with probability at least $1-\delta$}, whether the distributions are identical versus $\varepsilon$-far in total variation distance. Most prior work focused on the case that $\delta = \Omega(1)$, for which the sample complexity of identity testing is known to be $\Theta(\sqrt{n}/\epsilon^2)$. Given such an algorithm, one can achieve arbitrarily small values of $\delta$ via black-box amplification, which multiplies the required number of samples by $\Theta(\log(1/\delta))$. We show that black-box amplification is suboptimal for any $\delta = o(1)$, and give a new identity tester that achieves the optimal sample complexity. Our new upper and lower bounds show that the optimal sample complexity of identity testing is \[ \Theta\left( \frac{1}{\epsilon^2}\left(\sqrt{n \log(1/\delta)} + \log(1/\delta) \right)\right) \] for any $n, \varepsilon$, and $\delta$. For the special case of uniformity testing, where the given distribution is the uniform distribution $U_n$ over the domain, our new tester is surprisingly simple: to test whether $p = U_n$ versus $d_{\mathrm TV}(p, U_n) \geq \varepsilon$, we simply threshold $d_{\mathrm TV}(\widehat{p}, U_n)$, where $\widehat{p}$ is the empirical probability distribution. The fact that this simple "plug-in" estimator is sample-optimal is surprising, even in the constant $\delta$ case. Indeed, it was believed that such a tester would not attain sublinear sample complexity even for constant values of $\varepsilon$ and $\delta$.
  • We consider the problem of active linear regression with $\ell_2$-bounded noise. In this context, the learner receives a set of unlabeled data points, chooses a small subset to receive the labels for, and must give an estimate of the function that performs well on fresh samples. We give an algorithm that is simultaneously optimal in the number of labeled and unlabeled data points, with $O(d)$ labeled samples; previous work required $\Omega(d \log d)$ labeled samples regardless of the number of unlabeled samples. Our results also apply to learning linear functions from noisy queries, again achieving optimal sample complexities. The techniques extend beyond linear functions, giving improved sample complexities for learning the family of $k$-Fourier-sparse signals with continuous frequencies.
  • Multi-camera full-body pose capture of humans and animals in outdoor environments is a highly challenging problem. Our approach to it involves a team of cooperating micro aerial vehicles (MAVs) with on-board cameras only. The key enabling-aspect of our approach is the on-board person detection and tracking method. Recent state-of-the-art methods based on deep neural networks (DNN) are highly promising in this context. However, real time DNNs are severely constrained in input data dimensions, in contrast to available camera resolutions. Therefore, DNNs often fail at objects with small scale or far away from the camera, which are typical characteristics of a scenario with aerial robots. Thus, the core problem addressed in this paper is how to achieve on-board, real-time, continuous and accurate vision-based detections using DNNs for visual person tracking through MAVs. Our solution leverages cooperation among multiple MAVs. First, each MAV fuses its own detections with those obtained by other MAVs to perform cooperative visual tracking. This allows for predicting future poses of the tracked person, which are used to selectively process only the relevant regions of future images, even at high resolutions. Consequently, using our DNN-based detector we are able to continuously track even distant humans with high accuracy and speed. We demonstrate the efficiency of our approach through real robot experiments involving two aerial robots tracking a person, while maintaining an active perception-driven formation. Our solution runs fully on-board our MAV's CPU and GPU, with no remote processing. ROS-based source code is provided for the benefit of the community.
  • We consider the stochastic bandit problem in the sublinear space setting, where one cannot record the win-loss record for all $K$ arms. We give an algorithm using $O(1)$ words of space with regret \[ \sum_{i=1}^{K}\frac{1}{\Delta_i}\log \frac{\Delta_i}{\Delta}\log T \] where $\Delta_i$ is the gap between the best arm and arm $i$ and $\Delta$ is the gap between the best and the second-best arms. If the rewards are bounded away from $0$ and $1$, this is within an $O(\log 1/\Delta)$ factor of the optimum regret possible without space constraints.
  • We consider the problem of robust polynomial regression, where one receives samples $(x_i, y_i)$ that are usually within $\sigma$ of a polynomial $y = p(x)$, but have a $\rho$ chance of being arbitrary adversarial outliers. Previously, it was known how to efficiently estimate $p$ only when $\rho < \frac{1}{\log d}$. We give an algorithm that works for the entire feasible range of $\rho < 1/2$, while simultaneously improving other parameters of the problem. We complement our algorithm, which gives a factor 2 approximation, with impossibility results that show, for example, that a $1.09$ approximation is impossible even with infinitely many samples.
  • Sketching has emerged as a powerful technique for speeding up problems in numerical linear algebra, such as regression. In the overconstrained regression problem, one is given an $n \times d$ matrix $A$, with $n \gg d$, as well as an $n \times 1$ vector $b$, and one wants to find a vector $\hat{x}$ so as to minimize the residual error $\|Ax-b\|_2$. Using the sketch and solve paradigm, one first computes $S \cdot A$ and $S \cdot b$ for a randomly chosen matrix $S$, then outputs $x' = (SA)^{\dagger} Sb$ so as to minimize $\|SAx' - Sb\|_2$. The sketch-and-solve paradigm gives a bound on $\|x'-x^*\|_2$ when $A$ is well-conditioned. Our main result is that, when $S$ is the subsampled randomized Fourier/Hadamard transform, the error $x' - x^*$ behaves as if it lies in a "random" direction within this bound: for any fixed direction $a\in \mathbb{R}^d$, we have with $1 - d^{-c}$ probability that \[ \langle a, x'-x^*\rangle \lesssim \frac{\|a\|_2\|x'-x^*\|_2}{d^{\frac{1}{2}-\gamma}}, \quad (1) \] where $c, \gamma > 0$ are arbitrary constants. This implies $\|x'-x^*\|_{\infty}$ is a factor $d^{\frac{1}{2}-\gamma}$ smaller than $\|x'-x^*\|_2$. It also gives a better bound on the generalization of $x'$ to new examples: if rows of $A$ correspond to examples and columns to features, then our result gives a better bound for the error introduced by sketch-and-solve when classifying fresh examples. We show that not all oblivious subspace embeddings $S$ satisfy these properties. In particular, we give counterexamples showing that matrices based on Count-Sketch or leverage score sampling do not satisfy these properties. We also provide lower bounds, both on how small $\|x'-x^*\|_2$ can be, and for our new guarantee (1), showing that the subsampled randomized Fourier/Hadamard transform is nearly optimal.
  • The goal of compressed sensing is to estimate a vector from an underdetermined system of noisy linear measurements, by making use of prior knowledge on the structure of vectors in the relevant domain. For almost all results in this literature, the structure is represented by sparsity in a well-chosen basis. We show how to achieve guarantees similar to standard compressed sensing but without employing sparsity at all. Instead, we suppose that vectors lie near the range of a generative model $G: \mathbb{R}^k \to \mathbb{R}^n$. Our main theorem is that, if $G$ is $L$-Lipschitz, then roughly $O(k \log L)$ random Gaussian measurements suffice for an $\ell_2/\ell_2$ recovery guarantee. We demonstrate our results using generative models from published variational autoencoder and generative adversarial networks. Our method can use $5$-$10$x fewer measurements than Lasso for the same accuracy.
  • We study the problem of estimating the number of triangles in a graph stream. No streaming algorithm can get sublinear space on all graphs, so methods in this area bound the space in terms of parameters of the input graph such as the maximum number of triangles sharing a single edge. We give a sampling algorithm that is additionally parameterized by the maximum number of triangles sharing a single vertex. Our bound matches the best known turnstile results in all graphs, and gets better performance on simple graphs like $G(n, p)$ or a set of independent triangles. We complement the upper bound with a lower bound showing that no sampling algorithm can do better on those graphs by more than a log factor. In particular, any insertion stream algorithm must use $\sqrt{T}$ space when all the triangles share a common vertex, and any sampling algorithm must take $T^\frac{1}{3}$ samples when all the triangles are independent. We add another lower bound, also matching our algorithm's performance, which applies to all graph classes. This lower bound covers "triangle-dependent" sampling algorithms, a subclass that includes our algorithm and all previous sampling algorithms for the problem. Finally, we show how to generalize our algorithm to count arbitrary subgraphs of constant size.
  • We study the fundamental problems of (i) uniformity testing of a discrete distribution, and (ii) closeness testing between two discrete distributions with bounded $\ell_2$-norm. These problems have been extensively studied in distribution testing and sample-optimal estimators are known for them~\cite{Paninski:08, CDVV14, VV14, DKN:15}. In this work, we show that the original collision-based testers proposed for these problems ~\cite{GRdist:00, BFR+:00} are sample-optimal, up to constant factors. Previous analyses showed sample complexity upper bounds for these testers that are optimal as a function of the domain size $n$, but suboptimal by polynomial factors in the error parameter $\epsilon$. Our main contribution is a new tight analysis establishing that these collision-based testers are information-theoretically optimal, up to constant factors, both in the dependence on $n$ and in the dependence on $\epsilon$.
  • The Neural GPU is a recent model that can learn algorithms such as multi-digit binary addition and binary multiplication in a way that generalizes to inputs of arbitrary length. We show that there are two simple ways of improving the performance of the Neural GPU: by carefully designing a curriculum, and by increasing model size. The latter requires a memory efficient implementation, as a naive implementation of the Neural GPU is memory intensive. We find that these techniques increase the set of algorithmic problems that can be solved by the Neural GPU: we have been able to learn to perform all the arithmetic operations (and generalize to arbitrarily long numbers) when the arguments are given in the decimal representation (which, surprisingly, has not been possible before). We have also been able to train the Neural GPU to evaluate long arithmetic expressions with multiple operands that require respecting the precedence order of the operands, although these have succeeded only in their binary representation, and not with perfect accuracy. In addition, we gain insight into the Neural GPU by investigating its failure modes. We find that Neural GPUs that correctly generalize to arbitrarily long numbers still fail to compute the correct answer on highly-symmetric, atypical inputs: for example, a Neural GPU that achieves near-perfect generalization on decimal multiplication of up to 100-digit long numbers can fail on $000000\dots002 \times 000000\dots002$ while succeeding at $2 \times 2$. These failure modes are reminiscent of adversarial examples.
  • We propose a criterion for discrimination against a specified sensitive attribute in supervised learning, where the goal is to predict some target based on available features. Assuming data about the predictor, target, and membership in the protected group are available, we show how to optimally adjust any learned predictor so as to remove discrimination according to our definition. Our framework also improves incentives by shifting the cost of poor classification from disadvantaged groups to the decision maker, who can respond by improving the classification accuracy. In line with other studies, our notion is oblivious: it depends only on the joint statistics of the predictor, the target and the protected attribute, but not on interpretation of individualfeatures. We study the inherent limits of defining and identifying biases based on such oblivious measures, outlining what can and cannot be inferred from different oblivious tests. We illustrate our notion using a case study of FICO credit scores.
  • We consider the problem of estimating a Fourier-sparse signal from noisy samples, where the sampling is done over some interval $[0, T]$ and the frequencies can be "off-grid". Previous methods for this problem required the gap between frequencies to be above 1/T, the threshold required to robustly identify individual frequencies. We show the frequency gap is not necessary to estimate the signal as a whole: for arbitrary $k$-Fourier-sparse signals under $\ell_2$ bounded noise, we show how to estimate the signal with a constant factor growth of the noise and sample complexity polynomial in $k$ and logarithmic in the bandwidth and signal-to-noise ratio. As a special case, we get an algorithm to interpolate degree $d$ polynomials from noisy measurements, using $O(d)$ samples and increasing the noise by a constant factor in $\ell_2$.
  • In recent years, a number of works have studied methods for computing the Fourier transform in sublinear time if the output is sparse. Most of these have focused on the discrete setting, even though in many applications the input signal is continuous and naive discretization significantly worsens the sparsity level. We present an algorithm for robustly computing sparse Fourier transforms in the continuous setting. Let $x(t) = x^*(t) + g(t)$, where $x^*$ has a $k$-sparse Fourier transform and $g$ is an arbitrary noise term. Given sample access to $x(t)$ for some duration $T$, we show how to find a $k$-Fourier-sparse reconstruction $x'(t)$ with $$\frac{1}{T}\int_0^T |x'(t) - x(t) |^2 \mathrm{d} t \lesssim \frac{1}{T}\int_0^T | g(t)|^2 \mathrm{d}t.$$ The sample complexity is linear in $k$ and logarithmic in the signal-to-noise ratio and the frequency resolution. Previous results with similar sample complexities could not tolerate an infinitesimal amount of i.i.d. Gaussian noise, and even algorithms with higher sample complexities increased the noise by a polynomial factor. We also give new results for how precisely the individual frequencies of $x^*$ can be recovered.
  • We consider the problem of identifying the parameters of an unknown mixture of two arbitrary $d$-dimensional gaussians from a sequence of independent random samples. Our main results are upper and lower bounds giving a computationally efficient moment-based estimator with an optimal convergence rate, thus resolving a problem introduced by Pearson (1894). Denoting by $\sigma^2$ the variance of the unknown mixture, we prove that $\Theta(\sigma^{12})$ samples are necessary and sufficient to estimate each parameter up to constant additive error when $d=1.$ Our upper bound extends to arbitrary dimension $d>1$ up to a (provably necessary) logarithmic loss in $d$ using a novel---yet simple---dimensionality reduction technique. We further identify several interesting special cases where the sample complexity is notably smaller than our optimal worst-case bound. For instance, if the means of the two components are separated by $\Omega(\sigma)$ the sample complexity reduces to $O(\sigma^2)$ and this is again optimal. Our results also apply to learning each component of the mixture up to small error in total variation distance, where our algorithm gives strong improvements in sample complexity over previous work. We also extend our lower bound to mixtures of $k$ Gaussians, showing that $\Omega(\sigma^{6k-2})$ samples are necessary to estimate each parameter up to constant additive error.
  • We initiate the study of trade-offs between sparsity and the number of measurements in sparse recovery schemes for generic norms. Specifically, for a norm $\|\cdot\|$, sparsity parameter $k$, approximation factor $K>0$, and probability of failure $P>0$, we ask: what is the minimal value of $m$ so that there is a distribution over $m \times n$ matrices $A$ with the property that for any $x$, given $Ax$, we can recover a $k$-sparse approximation to $x$ in the given norm with probability at least $1-P$? We give a partial answer to this problem, by showing that for norms that admit efficient linear sketches, the optimal number of measurements $m$ is closely related to the doubling dimension of the metric induced by the norm $\|\cdot\|$ on the set of all $k$-sparse vectors. By applying our result to specific norms, we cast known measurement bounds in our general framework (for the $\ell_p$ norms, $p \in [1,2]$) as well as provide new, measurement-efficient schemes (for the Earth-Mover Distance norm). The latter result directly implies more succinct linear sketches for the well-studied planar $k$-median clustering problem. Finally, our lower bound for the doubling dimension of the EMD norm enables us to address the open question of [Frahling-Sohler, STOC'05] about the space complexity of clustering problems in the dynamic streaming model.
  • We provide a new robust convergence analysis of the well-known power method for computing the dominant singular vectors of a matrix that we call the noisy power method. Our result characterizes the convergence behavior of the algorithm when a significant amount noise is introduced after each matrix-vector multiplication. The noisy power method can be seen as a meta-algorithm that has recently found a number of important applications in a broad range of machine learning problems including alternating minimization for matrix completion, streaming principal component analysis (PCA), and privacy-preserving spectral analysis. Our general analysis subsumes several existing ad-hoc convergence bounds and resolves a number of open problems in multiple applications including streaming PCA and privacy-preserving singular vector computation.
  • Given a database, a common problem is to find the pairs or $k$-tuples of items that frequently co-occur. One specific problem is to create a small space "sketch" of the data that records which $k$-tuples appear in more than an $\epsilon$ fraction of rows of the database. We improve the lower bound of Liberty, Mitzenmacher, and Thaler [LMT14], showing that $\Omega(\frac{1}{\epsilon}d \log (\epsilon d))$ bits are necessary even in the case of $k=2$. This matches the sampling upper bound for all $\epsilon \geq 1/d^{.99}$, and (in the case of $k=2$) another trivial upper bound for $\epsilon = 1/d$.
  • We present a refined analysis of the classic Count-Sketch streaming heavy hitters algorithm [CCF02]. Count-Sketch uses O(k log n) linear measurements of a vector x in R^n to give an estimate x' of x. The standard analysis shows that this estimate x' satisfies ||x'-x||_infty^2 < ||x_tail||_2^2 / k, where x_tail is the vector containing all but the largest k coordinates of x. Our main result is that most of the coordinates of x' have substantially less error than this upper bound; namely, for any c < O(log n), we show that each coordinate i satisfies (x'_i - x_i)^2 < (c/log n) ||x_tail||_2^2/k with probability 1-2^{-Omega(c)}, as long as the hash functions are fully independent. This subsumes the previous bound and is optimal for all c. Using these improved point estimates, we prove a stronger concentration result for set estimates by first analyzing the covariance matrix and then using a median-of-median-of-medians argument to bootstrap the failure probability bounds. These results also give improved results for l_2 recovery of exactly k-sparse estimates x^* when x is drawn from a distribution with suitable decay, such as a power law or lognormal. We complement our results with simulations of Count-Sketch on a power law distribution. The empirical evidence indicates that our theoretical bounds give a precise characterization of the algorithm's performance: the asymptotics are correct and the associated constants are small. Our proof shows that any symmetric random variable with finite variance and positive Fourier transform concentrates around 0 at least as well as a Gaussian. This result, which may be of independent interest, gives good concentration even when the noise does not converge to a Gaussian.
  • We present the first sample-optimal sublinear time algorithms for the sparse Discrete Fourier Transform over a two-dimensional sqrt{n} x sqrt{n} grid. Our algorithms are analyzed for /average case/ signals. For signals whose spectrum is exactly sparse, our algorithms use O(k) samples and run in O(k log k) time, where k is the expected sparsity of the signal. For signals whose spectrum is approximately sparse, our algorithm uses O(k log n) samples and runs in O(k log^2 n) time; the latter algorithm works for k=Theta(sqrt{n}). The number of samples used by our algorithms matches the known lower bounds for the respective signal models. By a known reduction, our algorithms give similar results for the one-dimensional sparse Discrete Fourier Transform when n is a power of a small composite number (e.g., n = 6^t).
  • In compressed sensing, the "restricted isometry property" (RIP) is a sufficient condition for the efficient reconstruction of a nearly k-sparse vector x in C^d from m linear measurements Phi x. It is desirable for m to be small, and for Phi to support fast matrix-vector multiplication. In this work, we give a randomized construction of RIP matrices Phi in C^{m x d}, preserving the L_2 norms of all k-sparse vectors with distortion 1+eps, where the matrix-vector multiply Phi x can be computed in nearly linear time. The number of rows m is on the order of eps^{-2}klog dlog^2(klog d). Previous analyses of constructions of RIP matrices supporting fast matrix-vector multiplies, such as the sampled discrete Fourier matrix, required m to be larger by roughly a log k factor. Supporting fast matrix-vector multiplication is useful for iterative recovery algorithms which repeatedly multiply by Phi or Phi^*. Furthermore, our construction, together with a connection between RIP matrices and the Johnson-Lindenstrauss lemma in [Krahmer-Ward, SIAM. J. Math. Anal. 2011], implies fast Johnson-Lindenstrauss embeddings with asymptotically fewer rows than previously known. Our approach is a simple twist on previous constructions. Rather than choosing the rows for the embedding matrix to be rows sampled from some larger structured matrix (such as the discrete Fourier transform or a random circulant matrix), we instead choose each row of the embedding matrix to be a linear combination of a small number of rows of the original matrix, with random sign flips as coefficients. The main tool in our analysis is a recent bound for the supremum of certain types of Rademacher chaos processes in [Krahmer-Mendelson-Rauhut, arXiv:1207.0235].
  • We give lower bounds for the problem of stable sparse recovery from /adaptive/ linear measurements. In this problem, one would like to estimate a vector $x \in \R^n$ from $m$ linear measurements $A_1x,..., A_mx$. One may choose each vector $A_i$ based on $A_1x,..., A_{i-1}x$, and must output $x*$ satisfying |x* - x|_p \leq (1 + \epsilon) \min_{k\text{-sparse} x'} |x - x'|_p with probability at least $1-\delta>2/3$, for some $p \in \{1,2\}$. For $p=2$, it was recently shown that this is possible with $m = O(\frac{1}{\epsilon}k \log \log (n/k))$, while nonadaptively it requires $\Theta(\frac{1}{\epsilon}k \log (n/k))$. It is also known that even adaptively, it takes $m = \Omega(k/\epsilon)$ for $p = 2$. For $p = 1$, there is a non-adaptive upper bound of $\tilde{O}(\frac{1}{\sqrt{\epsilon}} k\log n)$. We show: * For $p=2$, $m = \Omega(\log \log n)$. This is tight for $k = O(1)$ and constant $\epsilon$, and shows that the $\log \log n$ dependence is correct. * If the measurement vectors are chosen in $R$ "rounds", then $m = \Omega(R \log^{1/R} n)$. For constant $\epsilon$, this matches the previously known upper bound up to an O(1) factor in $R$. * For $p=1$, $m = \Omega(k/(\sqrt{\epsilon} \cdot \log k/\epsilon))$. This shows that adaptivity cannot improve more than logarithmic factors, providing the analog of the $m = \Omega(k/\epsilon)$ bound for $p = 2$.
  • We initiate the study of sparse recovery problems under the Earth-Mover Distance (EMD). Specifically, we design a distribution over m x n matrices A such that for any x, given Ax, we can recover a k-sparse approximation to x under the EMD distance. One construction yields m = O(k log(n/k)) and a 1 + epsilon approximation factor, which matches the best achievable bound for other error measures, such as the L_1 norm. Our algorithms are obtained by exploiting novel connections to other problems and areas, such as streaming algorithms for k-median clustering and model-based compressive sensing. We also provide novel algorithms and results for the latter problems.
  • We propose a framework for compressive sensing of images with local distinguishable objects, such as stars, and apply it to solve a problem in celestial navigation. Specifically, let x be an N-pixel real-valued image, consisting of a small number of local distinguishable objects plus noise. Our goal is to design an m-by-N measurement matrix A with m << N, such that we can recover an approximation to x from the measurements Ax. We construct a matrix A and recovery algorithm with the following properties: (i) if there are k objects, the number of measurements m is O((k log N)/(log k)), undercutting the best known bound of O(k log(N/k)) (ii) the matrix A is very sparse, which is important for hardware implementations of compressive sensing algorithms, and (iii) the recovery algorithm is empirically fast and runs in time polynomial in k and log(N). We also present a comprehensive study of the application of our algorithm to attitude determination, or finding one's orientation in space. Spacecraft typically use cameras to acquire an image of the sky, and then identify stars in the image to compute their orientation. Taking pictures is very expensive for small spacecraft, since camera sensors use a lot of power. Our algorithm optically compresses the image before it reaches the camera's array of pixels, reducing the number of sensors that are required.
  • We consider the problem of computing the k-sparse approximation to the discrete Fourier transform of an n-dimensional signal. We show: * An O(k log n)-time randomized algorithm for the case where the input signal has at most k non-zero Fourier coefficients, and * An O(k log n log(n/k))-time randomized algorithm for general input signals. Both algorithms achieve o(n log n) time, and thus improve over the Fast Fourier Transform, for any k = o(n). They are the first known algorithms that satisfy this property. Also, if one assumes that the Fast Fourier Transform is optimal, the algorithm for the exactly k-sparse case is optimal for any k = n^{\Omega(1)}. We complement our algorithmic results by showing that any algorithm for computing the sparse Fourier transform of a general signal must use at least \Omega(k log(n/k)/ log log n) signal samples, even if it is allowed to perform adaptive sampling.