arXiv subjects

Parikshit Gopalan

Publications and source records attributed to Parikshit Gopalan.

At least 37 records · Page 2Linked to original sources

arXiv2022

Low-Degree Multicalibration

Introduced as a notion of algorithmic fairness, multicalibration has proved to be a powerful and versatile concept with implications far beyond its original intent. This stringent notion -- that predictions be well-calibrated across a rich class of intersecting subpopulations -- provides its strong guarantees at a cost: the computational and sample complexity of learning multicalibrated predictors are high, and grow exponentially with the number of class labels. In contrast, the relaxed notion of multiaccuracy can be achieved more efficiently, yet many of the most desirable properties of multicalibration cannot be guaranteed assuming multiaccuracy alone. This tension raises a key question: Can we learn predictors with multicalibration-style guarantees at a cost commensurate with multiaccuracy? In this work, we define and initiate the study of Low-Degree Multicalibration. Low-Degree Multicalibration defines a hierarchy of increasingly-powerful multi-group fairness notions that spans multiaccuracy and the original formulation of multicalibration at the extremes. Our main technical contribution demonstrates that key properties of multicalibration, related to fairness and accuracy, actually manifest as low-degree properties. Importantly, we show that low-degree multicalibration can be significantly more efficient than full multicalibration. In the multi-class setting, the sample complexity to achieve low-degree multicalibration improves exponentially (in the number of classes) over full multicalibration. Our work presents compelling evidence that low-degree multicalibration represents a sweet spot, pairing computational and sample efficiency with strong fairness and accuracy guarantees.

cs.LG↗

arXiv2022

KL Divergence Estimation with Multi-group Attribution

Estimating the Kullback-Leibler (KL) divergence between two distributions given samples from them is well-studied in machine learning and information theory. Motivated by considerations of multi-group fairness, we seek KL divergence estimates that accurately reflect the contributions of sub-populations to the overall divergence. We model the sub-populations coming from a rich (possibly infinite) family $\mathcal{C}$ of overlapping subsets of the domain. We propose the notion of multi-group attribution for $\mathcal{C}$, which requires that the estimated divergence conditioned on every sub-population in $\mathcal{C}$ satisfies some natural accuracy and fairness desiderata, such as ensuring that sub-populations where the model predicts significant divergence do diverge significantly in the two distributions. Our main technical contribution is to show that multi-group attribution can be derived from the recently introduced notion of multi-calibration for importance weights [HKRR18, GRSW21]. We provide experimental evidence to support our theoretical results, and show that multi-group attribution provides better KL divergence estimates when conditioned on sub-populations than other popular algorithms.

cs.LG↗

arXiv2021

Omnipredictors

Loss minimization is a dominant paradigm in machine learning, where a predictor is trained to minimize some loss function that depends on an uncertain event (e.g., "will it rain tomorrow?''). Different loss functions imply different learning algorithms and, at times, very different predictors. While widespread and appealing, a clear drawback of this approach is that the loss function may not be known at the time of learning, requiring the algorithm to use a best-guess loss function. We suggest a rigorous new paradigm for loss minimization in machine learning where the loss function can be ignored at the time of learning and only be taken into account when deciding an action. We introduce the notion of an (${\mathcal{L}},\mathcal{C}$)-omnipredictor, which could be used to optimize any loss in a family ${\mathcal{L}}$. Once the loss function is set, the outputs of the predictor can be post-processed (a simple univariate data-independent transformation of individual predictions) to do well compared with any hypothesis from the class $\mathcal{C}$. The post processing is essentially what one would perform if the outputs of the predictor were true probabilities of the uncertain events. In a sense, omnipredictors extract all the predictive power from the class $\mathcal{C}$, irrespective of the loss function in $\mathcal{L}$. We show that such "loss-oblivious'' learning is feasible through a connection to multicalibration, a notion introduced in the context of algorithmic fairness. In addition, we show how multicalibration can be viewed as a solution concept for agnostic boosting, shedding new light on past results. Finally, we transfer our insights back to the context of algorithmic fairness by providing omnipredictors for multi-group loss minimization.

cs.LG↗

arXiv2021

Multicalibrated Partitions for Importance Weights

The ratio between the probability that two distributions $R$ and $P$ give to points $x$ are known as importance weights or propensity scores and play a fundamental role in many different fields, most notably, statistics and machine learning. Among its applications, importance weights are central to domain adaptation, anomaly detection, and estimations of various divergences such as the KL divergence. We consider the common setting where $R$ and $P$ are only given through samples from each distribution. The vast literature on estimating importance weights is either heuristic, or makes strong assumptions about $R$ and $P$ or on the importance weights themselves. In this paper, we explore a computational perspective to the estimation of importance weights, which factors in the limitations and possibilities obtainable with bounded computational resources. We significantly strengthen previous work that use the MaxEntropy approach, that define the importance weights based on a distribution $Q$ closest to $P$, that looks the same as $R$ on every set $C \in \mathcal{C}$, where $\mathcal{C}$ may be a huge collection of sets. We show that the MaxEntropy approach may fail to assign high average scores to sets $C \in \mathcal{C}$, even when the average of ground truth weights for the set is evidently large. We similarly show that it may overestimate the average scores to sets $C \in \mathcal{C}$. We therefore formulate Sandwiching bounds as a notion of set-wise accuracy for importance weights. We study these bounds to show that they capture natural completeness and soundness requirements from the weights. We present an efficient algorithm that under standard learnability assumptions computes weights which satisfy these bounds. Our techniques rely on a new notion of multicalibrated partitions of the domain of the distributions, which appear to be useful objects in their own right.

cs.LG↗

arXiv2020

Finding Skewed Subcubes Under a Distribution

Say that we are given samples from a distribution $ψ$ over an $n$-dimensional space. We expect or desire $ψ$ to behave like a product distribution (or a $k$-wise independent distribution over its marginals for small $k$). We propose the problem of enumerating/list-decoding all large subcubes where the distribution $ψ$ deviates markedly from what we expect; we refer to such subcubes as skewed subcubes. Skewed subcubes are certificates of dependencies between small subsets of variables in $ψ$. We motivate this problem by showing that it arises naturally in the context of algorithmic fairness and anomaly detection. In this work we focus on the special but important case where the space is the Boolean hypercube, and the expected marginals are uniform. We show that the obvious definition of skewed subcubes can lead to intractable list sizes, and propose a better definition of a minimal skewed subcube, which are subcubes whose skew cannot be attributed to a larger subcube that contains it. Our main technical contribution is a list-size bound for this definition and an algorithm to efficiently find all such subcubes. Both the bound and the algorithm rely on Fourier-analytic techniques, especially the powerful hypercontractive inequality. On the lower bounds side, we show that finding skewed subcubes is as hard as the sparse noisy parity problem, and hence our algorithms cannot be improved on substantially without a breakthrough on this problem which is believed to be intractable. Motivated by this, we study alternate models allowing query access to $ψ$ where finding skewed subcubes might be easier.

cs.DS↗

arXiv2020

Overlook: Differentially Private Exploratory Visualization for Big Data

Data exploration systems that provide differential privacy must manage a privacy budget that measures the amount of privacy lost across multiple queries. One effective strategy to manage the privacy budget is to compute a one-time private synopsis of the data, to which users can make an unlimited number of queries. However, existing systems using synopses are built for offline use cases, where a set of queries is known ahead of time and the system carefully optimizes a synopsis for it. The synopses that these systems build are costly to compute and may also be costly to store. We introduce Overlook, a system that enables private data exploration at interactive latencies for both data analysts and data curators. The key idea in Overlook is a virtual synopsis that can be evaluated incrementally, without extra space storage or expensive precomputation. Overlook simply executes queries using an existing engine, such as a SQL DBMS, and adds noise to their results. Because Overlook's synopses do not require costly precomputation or storage, data curators can also use Overlook to explore the impact of privacy parameters interactively. Overlook offers a rich visual query interface based on the open source Hillview system. Overlook achieves accuracy comparable to existing synopsis-based systems, while offering better performance and removing the need for extra storage.

cs.CR↗

arXiv2019

PIDForest: Anomaly Detection via Partial Identification

We consider the problem of detecting anomalies in a large dataset. We propose a framework called Partial Identification which captures the intuition that anomalies are easy to distinguish from the overwhelming majority of points by relatively few attribute values. Formalizing this intuition, we propose a geometric anomaly measure for a point that we call PIDScore, which measures the minimum density of data points over all subcubes containing the point. We present PIDForest: a random forest based algorithm that finds anomalies based on this definition. We show that it performs favorably in comparison to several popular anomaly detection methods, across a broad range of benchmarks. PIDForest also provides a succinct explanation for why a point is labelled anomalous, by providing a set of features and ranges for them which are relatively uncommon in the dataset.

cs.LG↗

arXiv2019

Hillview: A trillion-cell spreadsheet for big data

Hillview is a distributed spreadsheet for browsing very large datasets that cannot be handled by a single machine. As a spreadsheet, Hillview provides a high degree of interactivity that permits data analysts to explore information quickly along many dimensions while switching visualizations on a whim. To provide the required responsiveness, Hillview introduces visualization sketches, or vizketches, as a simple idea to produce compact data visualizations. Vizketches combine algorithmic techniques for data summarization with computer graphics principles for efficient rendering. While simple, vizketches are effective at scaling the spreadsheet by parallelizing computation, reducing communication, providing progressive visualizations, and offering precise accuracy guarantees. Using Hillview running on eight servers, we can navigate and visualize datasets of tens of billions of rows and trillions of cells, much beyond the published capabilities of competing systems.

cs.DC↗

arXiv2018

Efficient Anomaly Detection via Matrix Sketching

We consider the problem of finding anomalies in high-dimensional data using popular PCA based anomaly scores. The naive algorithms for computing these scores explicitly compute the PCA of the covariance matrix which uses space quadratic in the dimensionality of the data. We give the first streaming algorithms that use space that is linear or sublinear in the dimension. We prove general results showing that \emph{any} sketch of a matrix that satisfies a certain operator norm guarantee can be used to approximate these scores. We instantiate these results with powerful matrix sketching techniques such as Frequent Directions and random projections to derive efficient and practical algorithms for these problems, which we validate over real-world data sets. Our main technical contribution is to prove matrix perturbation inequalities for operators arising in the computation of these measures.

cs.LG↗

arXiv2018

Stable and Consistent Membership at Scale with Rapid

We present the design and evaluation of Rapid, a distributed membership service. At Rapid's core is a scheme for multi-process cut detection (CD) that revolves around two key insights: (i) it suspects a failure of a process only after alerts arrive from multiple sources, and (ii) when a group of processes experience problems, it detects failures of the entire group, rather than conclude about each process individually. Implementing these insights translates into a simple membership algorithm with low communication overhead. We present evidence that our strategy suffices to drive unanimous detection almost-everywhere, even when complex network conditions arise, such as one-way reachability problems, firewall misconfigurations, and high packet loss. Furthermore, we present both empirical evidence and analyses that proves that the almost-everywhere detection happens with high probability. To complete the design, Rapid contains a leaderless consensus protocol that converts multi-process cut detections into a view-change decision. The resulting membership service works both in fully decentralized as well as logically centralized modes. We present an evaluation of Rapid in moderately scalable cloud settings. Rapid bootstraps 2000 node clusters 2-5.8x faster than prevailing tools such as Memberlist and ZooKeeper, remains stable in face of complex failure scenarios, and provides strong consistency guarantees. It is easy to integrate Rapid into existing distributed applications, of which we demonstrate two.

cs.DC↗

arXiv2016

Maximally Recoverable Codes for Grid-like Topologies

The explosion in the volumes of data being stored online has resulted in distributed storage systems transitioning to erasure coding based schemes. Yet, the codes being deployed in practice are fairly short. In this work, we address what we view as the main coding theoretic barrier to deploying longer codes in storage: at large lengths, failures are not independent and correlated failures are inevitable. This motivates designing codes that allow quick data recovery even after large correlated failures, and which have efficient encoding and decoding. We propose that code design for distributed storage be viewed as a two-step process. The first step is choose a topology of the code, which incorporates knowledge about the correlated failures that need to be handled, and ensures local recovery from such failures. In the second step one specifies a code with the chosen topology by choosing coefficients from a finite field. In this step, one tries to balance reliability (which is better over larger fields) with encoding and decoding efficiency (which is better over smaller fields). This work initiates an in-depth study of this reliability/efficiency tradeoff. We consider the field-size needed for achieving maximal recoverability: the strongest reliability possible with a given topology. We propose a family of topologies called grid-like topologies which unify a number of topologies considered both in theory and practice, and prove a collection of results about maximally recoverable codes in such topologies including the first super-polynomial lower bound on the field size.

cs.IT↗

arXiv2016

Degree and Sensitivity: tails of two distributions

The sensitivity of a Boolean function f is the maximum over all inputs x, of the number of sensitive coordinates of x. The well-known sensitivity conjecture of Nisan (see also Nisan and Szegedy) states that every sensitivity-s Boolean function can be computed by a polynomial over the reals of degree poly(s). The best known upper bounds on degree, however, are exponential rather than polynomial in s. Our main result is an approximate version of the conjecture: every Boolean function with sensitivity s can be epsilon-approximated (in L_2) by a polynomial whose degree is O(s log(1/epsilon)). This is the first improvement on the folklore bound of s/epsilon. Further, we show that improving the bound to O(s^c log(1/epsilon)^d)$ for any d < 1 and any c > 0 will imply the sensitivity conjecture. Thus our result is essentially the best one can hope for without proving the conjecture. We postulate a robust analogue of the sensitivity conjecture: if most inputs to a Boolean function f have low sensitivity, then most of the Fourier mass of f is concentrated on small subsets, and present an approach towards proving this conjecture.

cs.CC↗

arXiv2015

Pseudorandomness via the discrete Fourier transform

We present a new approach to constructing unconditional pseudorandom generators against classes of functions that involve computing a linear function of the inputs. We give an explicit construction of a pseudorandom generator that fools the discrete Fourier transforms of linear functions with seed-length that is nearly logarithmic (up to polyloglog factors) in the input size and the desired error parameter. Our result gives a single pseudorandom generator that fools several important classes of tests computable in logspace that have been considered in the literature, including halfspaces (over general domains), modular tests and combinatorial shapes. For all these classes, our generator is the first that achieves near logarithmic seed-length in both the input length and the error parameter. Getting such a seed-length is a natural challenge in its own right, which needs to be overcome in order to derandomize RL - a central question in complexity theory. Our construction combines ideas from a large body of prior work, ranging from a classical construction of [NN93] to the recent gradually increasing independence paradigm of [KMN11, CRSW13, GMRTV12], while also introducing some novel analytic machinery which might find other applications.

cs.CC↗

arXiv2015

Inequalities and tail bounds for elementary symmetric polynomial with applications

We study the extent of independence needed to approximate the product of bounded random variables in expectation, a natural question that has applications in pseudorandomness and min-wise independent hashing. For random variables whose absolute value is bounded by $1$, we give an error bound of the form $σ^{Ω(k)}$ where $k$ is the amount of independence and $σ^2$ is the total variance of the sum. Previously known bounds only applied in more restricted settings, and were quanitively weaker. We use this to give a simpler and more modular analysis of a construction of min-wise independent hash functions and pseudorandom generators for combinatorial rectangles due to Gopalan et al., which also slightly improves their seed-length. Our proof relies on a new analytic inequality for the elementary symmetric polynomials $S_k(x)$ for $x \in \mathbb{R}^n$ which we believe to be of independent interest. We show that if $|S_k(x)|,|S_{k+1}(x)|$ are small relative to $|S_{k-1}(x)|$ for some $k>0$ then $|S_\ell(x)|$ is also small for all $\ell > k$. From these, we derive tail bounds for the elementary symmetric polynomials when the inputs are only $k$-wise independent.

cs.CC↗

arXiv2015

Smooth Boolean functions are easy: efficient algorithms for low-sensitivity functions

A natural measure of smoothness of a Boolean function is its sensitivity (the largest number of Hamming neighbors of a point which differ from it in function value). The structure of smooth or equivalently low-sensitivity functions is still a mystery. A well-known conjecture states that every such Boolean function can be computed by a shallow decision tree. While this conjecture implies that smooth functions are easy to compute in the simplest computational model, to date no non-trivial upper bounds were known for such functions in any computational model, including unrestricted Boolean circuits. Even a bound on the description length of such functions better than the trivial $2^n$ does not seem to have been known. In this work, we establish the first computational upper bounds on smooth Boolean functions: 1) We show that every sensitivity s function is uniquely specified by its values on a Hamming ball of radius 2s. We use this to show that such functions can be computed by circuits of size $n^{O(s)}$. 2) We show that sensitivity s functions satisfy a strong pointwise noise-stability guarantee for random noise of rate O(1/s). We use this to show that these functions have formulas of depth O(s log n). 3) We show that sensitivity s functions can be (locally) self-corrected from worst-case noise of rate $\exp(-O(s))$. All our results are simple, and follow rather directly from (variants of) the basic fact that that the function value at few points in small neighborhoods of a given point determine its function value via a majority vote. Our results confirm various consequences of the conjecture. They may be viewed as providing a new form of evidence towards its validity, as well as new directions towards attacking it.

cs.CC↗

arXiv2015

Public projects, Boolean functions and the borders of Border's theorem

Border's theorem gives an intuitive linear characterization of the feasible interim allocation rules of a Bayesian single-item environment, and it has several applications in economic and algorithmic mechanism design. All known generalizations of Border's theorem either restrict attention to relatively simple settings, or resort to approximation. This paper identifies a complexity-theoretic barrier that indicates, assuming standard complexity class separations, that Border's theorem cannot be extended significantly beyond the state-of-the-art. We also identify a surprisingly tight connection between Myerson's optimal auction theory, when applied to public project settings, and some fundamental results in the analysis of Boolean functions.

cs.GT↗

arXiv2014

Pseudorandomness for concentration bounds and signed majorities

The problem of constructing pseudorandom generators that fool halfspaces has been studied intensively in recent times. For fooling halfspaces over the hypercube with polynomially small error, the best construction known requires seed-length O(log^2 n) (MekaZ13). Getting the seed-length down to O(log(n)) is a natural challenge in its own right, which needs to be overcome in order to derandomize RL. In this work we make progress towards this goal by obtaining near-optimal generators for two important special cases: 1) We give a near optimal derandomization of the Chernoff bound for independent, uniformly random bits. Specifically, we show how to generate a x in {1,-1}^n using $\tilde{O}(\log (n/ε))$ random bits such that for any unit vector u, matches the sub-Gaussian tail behaviour predicted by the Chernoff bound up to error eps. 2) We construct a generator which fools halfspaces with {0,1,-1} coefficients with error eps with a seed-length of $\tilde{O}(\log(n/ε))$. This includes the important special case of majorities. In both cases, the best previous results required seed-length of $O(\log n + \log^2(1/ε))$. Technically, our work combines new Fourier-analytic tools with the iterative dimension reduction techniques and the gradually increasing independence paradigm of previous works (KaneMN11, CelisRSW13, GopalanMRTV12).

cs.CC↗

arXiv2013

Locally Testable Codes and Cayley Graphs

We give two new characterizations of ($\F_2$-linear) locally testable error-correcting codes in terms of Cayley graphs over $\F_2^h$: \begin{enumerate} \item A locally testable code is equivalent to a Cayley graph over $\F_2^h$ whose set of generators is significantly larger than $h$ and has no short linear dependencies, but yields a shortest-path metric that embeds into $\ell_1$ with constant distortion. This extends and gives a converse to a result of Khot and Naor (2006), which showed that codes with large dual distance imply Cayley graphs that have no low-distortion embeddings into $\ell_1$. \item A locally testable code is equivalent to a Cayley graph over $\F_2^h$ that has significantly more than $h$ eigenvalues near 1, which have no short linear dependencies among them and which "explain" all of the large eigenvalues. This extends and gives a converse to a recent construction of Barak et al. (2012), which showed that locally testable codes imply Cayley graphs that are small-set expanders but have many large eigenvalues. \end{enumerate}

cs.CC↗