Searcharxiv⌕ Search

arXiv subjects

Amir Yehudayoff

Publications and source records attributed to Amir Yehudayoff.

At least 19 recordsLinked to original sources

Derandomized Sunflowers and Radical Lower Bounds

Robust Sunflower lemmas imply that any large enough monotone DNF of width $w$ contains a sunflower, i.e., a DNF equivalent to the conjunction of a common core with a DNF that is heavily biased towards $1$. While these lemmas are typically proved in the context of the uniform distribution or i.i.d. distributions, they also hold for $O(w^2)$-wise independent distributions by Bazzi's theorem (Bazzi (SICOMP 2009); Tal (CCC 2017)), which states that width-$w$ DNFs cannot distinguish between such distributions and the uniform distribution. We prove a derandomization of robust sunflower lemmas for $O(w)$-wise independent distributions, which recovers the derandomization above with slightly weaker parameters. Notably, our result also holds for distributions whose marginals are "locally independent" and "approximately identically" distributed. It is not known whether such distributions fool general DNFs, which makes the derandomization via DNF-fooling techniques unavailable. Our proof of this derandomization is via the method of moments, which might be of independent interest. We apply this result to obtain improved separations between non-monotone arithmetic circuits and the radical of monotone circuits. Specifically, we construct an explicit family of multilinear polynomials $P_n$ that have small non-monotone arithmetic circuits such that any power of $P_n$ cannot be computed by monotone arithmetic circuits of sub-exponential size. This improves on a recent result of Cavalar, Fabris, Mukhopadhyay, Srinivasan and Yehudayoff (STOC 2026), who proved a non-explicit separation that was quasipolynomial. Our construction uses Nisan-Wigderson designs with an additional linear-algebraic expansion property.

cs.CC↗

On the Depth of Monotone ReLU Neural Networks and ICNNs

We study two models of $\mathsf{ReLU}$ neural networks: monotone networks ($\mathsf{ReLU}^+$) and input convex neural networks ($\mathsf{ICNN}$). Our focus is on expressivity, mostly in terms of depth, and our motivation is to gain a better understanding of the depth required to exactly represent functions using neural networks with ReLU activations, a subject that has received a lot of attention lately. We prove several lower bounds on the depth required to represent functions by monotone ReLU networks and $\mathsf{ICNN}$. For the maximum function $\mathsf{MAX}_n$ computing the maximum of $n$ real numbers, we show that $\mathsf{ReLU}^+$ networks cannot compute $\mathsf{MAX}_n$, or even approximate it. We prove a sharp lower bound of $n-1$ on the $\mathsf{ICNN}$ depth complexity of $\mathsf{MAX}_n$. We also prove depth separations between $\mathsf{ReLU}$ networks and $\mathsf{ICNN}$s; for every $k$, there is a depth-$2$ $\mathsf{ReLU}$ network of size $O(k^2)$ that cannot be simulated by a depth-$k$ $\mathsf{ICNN}$. The proofs are based on deep connections between neural networks and polyhedral geometry, and also use isoperimetric properties of triangulations.

cs.LG↗

Two $\operatorname{ReLU}$ Layers is All You Need

We prove that every $\operatorname{ReLU}$ network can be exactly simulated by a $\operatorname{ReLU}$ network with two hidden layers. We present two proofs: one that is based on fillings in homological algebra and a direct combinatorial proof.

cs.CC↗

Volume stability of plank covers

We prove a longstanding conjecture of András Bezdek on the stability of plank covers of the planar disk $D$. If a sufficiently small disk $\varepsilon D$ is removed, then every finite family of planks covering the resulting annulus can be rearranged to cover the whole disk. For the proof, we establish a stronger dimension-independent stability theorem. If $K \subset {\mathbb R}^n$ is an origin-symmetric convex body for $n \geq 2$, and the uncovered region lies in $\varepsilon K$ and has volume $v$, then the excess overlap is at least $c v/\varepsilon$, where $c>0$ is an absolute constant. The proof introduces a pruning procedure and a flattening mechanism. Pruning peels away curvature through local removals whose cumulative effect is guaranteed to be substantial. Flattening trades straightness for bounded multiplicity and geometrically exposes the overlap that pays for each pruning step.

math.MG↗

A note on shifting distributions via Poisson races

This expository note is about simulating a target distribution $Q$ from observations of a proposal distribution $P$. In the model suggested by Harsha, Jain, McAllester and Radhakrishnan, we observe an infinite sequence of i.i.d. samples $X_1,X_2,\ldots$ from $P$. The goal is to find some index $I \in \{1,2,\ldots\}$ such that $X_I$ is distributed like $Q$ while minimizing $\mathbb{E} \log I$. Following the Poisson-race approach developed by Maddison, Li and El Gamal, and others, this note shows that if $D(Q||P) < \infty$ then there is a $P$ to $Q$ simulator $I$ such that $\mathbb{E} [\log I] \leq D(Q||P) + 1.45 \|Q-P\|_1$. In the other direction, for every $P$ to $Q$ simulator $I$, the cost is at least $\mathbb{E}[\log I] \geq \frac{1}{2} \max \{ D(Q||P), \|Q-P\|_1 \}$.

math.PR↗

A note on the affine plank conjecture

In 1951, Bang posed the affine plank conjecture, which remains open: If a convex body in $\mathbb{R}^d$ is covered by planks, then the total relative width of the planks is at least one. We prove a lower bound of $2/(1+\sqrt{d})$ for this total relative width. The best previously known lower bound was $2/(1+d)$.

math.MG↗

A simplex-based measure of symmetry

For compact convex sets $L,K \subset \mathbb{R}^n$, denote by $λ_K(L)$ the smallest size of a homothet of $K$ that contains $L$. We define a measure of symmetry based on the $n$-simplex $Δ= Δ^n \subset \mathbb{R}^n$ as the ratio \[ ρ_Δ(L):=\frac{λ_{-Δ}(L)}{λ_Δ(L)}. \] We study this measure and deduce the following results: (1) The classical Minkowski measure of symmetry $m^*(L)$ can be defined as an affine-invariant version of $ρ_Δ(L)$. (2) We improve the stability analysis for the Minkowski measure of symmetry; if $m^*(L)\ge n-\varepsilon$ then $L$ is $\tfrac{1}{1-\varepsilon}$-close to $Δ$ in the Banach--Mazur distance. (3) We obtain a novel characterization of simplices as the only convex bodies $K$ for which the function $L \mapsto λ_K(L)$ is additive (a property we term ``outer additivity''). (4) Motivated by the expressivity of ReLU neural networks, we study the depth complexity of polytopes in $\mathbb{R}^n$ under the two operations: Minkowski sum and convex hull of a union. We prove the sharp bound $ρ_Δ(P) \leq 2^d -1$ for every polytope $P$ of depth complexity $d$. In other words, simplices cannot be approximated by low-depth polytopes.

math.MG↗

Better Neural Network Expressivity: Subdividing the Simplex

This work studies the expressivity of ReLU neural networks with a focus on their depth. A sequence of previous works showed that $\lceil \log_2(n+1) \rceil$ hidden layers are sufficient to compute all continuous piecewise linear (CPWL) functions on $\mathbb{R}^n$. Hertrich, Basu, Di Summa, and Skutella (NeurIPS'21 / SIDMA'23) conjectured that this result is optimal in the sense that there are CPWL functions on $\mathbb{R}^n$, like the maximum function, that require this depth. We disprove the conjecture and show that $\lceil\log_3(n-1)\rceil+1$ hidden layers are sufficient to compute all CPWL functions on $\mathbb{R}^n$. A key step in the proof is that ReLU neural networks with two hidden layers can exactly represent the maximum function of five inputs. More generally, we show that $\lceil\log_3(n-2)\rceil+1$ hidden layers are sufficient to compute the maximum of $n\geq 4$ numbers. Our constructions almost match the $\lceil\log_3(n)\rceil$ lower bound of Averkov, Hojny, and Merkert (ICLR'25) in the special case of ReLU networks with weights that are decimal fractions. The constructions have a geometric interpretation via polyhedral subdivisions of the simplex into ``easier'' polytopes.

cs.LG↗

On the Space Complexity of Online Convolution

We study a discrete convolution streaming problem. An input arrives as a stream of numbers $z = (z_0,z_1,z_2,\ldots)$, and at time $t$ our goal is to output $(Tz)_t$ where $T$ is a lower-triangular Toeplitz matrix. We focus on space complexity; we define a model for studying the memory-size of online continuous algorithms. In this model, algorithms store a buffer of $β(t)$ numbers in order to achieve their goal. We characterize space complexity using the language of generating functions. The matrix $T$ corresponds to a generating function $G(x)$. When $G(x)$ is rational of degree $d$, it is known that the space complexity is at most $O(d)$. We prove a corresponding lower bound; the space complexity is at least $Ω(d)$. In addition, for irrational $G(x)$, we prove that the space complexity is infinite. We also provide finite-time guarantees. For example, for the generating function $\frac{1}{\sqrt{1-x}}$ that was studied in various previous works in the context of differentially private continual counting, we prove a sharp lower bound on the space complexity; at time $t$, it is at least $Ω(t)$.

cs.CC↗

Negations are powerful even in small depth

We study the power of negation in the Boolean and algebraic settings and show the following results. * We construct a family of polynomials $P_n$ in $n$ variables, all of whose monomials have positive coefficients, such that $P_n$ can be computed by a depth three circuit of polynomial size but any monotone circuit computing it has size $2^{Ω(n)}$. This is the strongest possible separation result between monotone and non-monotone arithmetic computations and improves upon all earlier results, including the seminal work of Valiant (1980) and more recently by Chattopadhyay, Datta, and Mukhopadhyay (2021). We then boot-strap this result to prove strong monotone separations for polynomials of constant degree, which solves an open problem from the survey of Shpilka and Yehudayoff (2010). * By moving to the Boolean setting, we can prove superpolynomial monotone Boolean circuit lower bounds for specific Boolean functions, which imply that all the powers of certain monotone polynomials cannot be computed by polynomially sized monotone arithmetic circuits. * We then define a collection of problems with linear-algebraic nature, which are similar to span programs, and prove monotone Boolean circuit lower bounds for them. In particular, this gives the strongest known monotone lower bounds for functions in uniform (non-monotone) $\textbf{NC}^2$. Our construction also leads to an explicit matroid that defines a monotone function that is difficult to compute, which solves an open problem by Jukna and Seiwert (2020). Our monotone arithmetic and Boolean circuit lower bounds are based on known techniques, such as reduction from monotone arithmetic complexity to multipartition communication complexity and the approximation method for proving lower bounds for monotone Boolean circuits, but we overcome several new challenges in order to obtain efficient upper bounds using low-depth circuits.

cs.CC↗

A Blaschke-Santaló inequality for unconditional log-concave measures

The Blaschke-Santaló inequality states that the volume product $|K| \cdot |K^{o}|$ of a symmetric convex body $K \subset \mathbb{R}^n$ is maximized by the standard Euclidean unit-ball. Cordero-Erausquin asked whether the inequality remains true for all even log-concave measures. We briefly survey the literature around this question and provide details for the known fact that the inequality holds true for all unconditional log-concave measures.

math.FA↗

The Algebraic Cost of a Boolean Sum

It is a well-known fact that the permanent polynomial is complete for the complexity class VNP, and it is largely suspected that the determinant does not share this property, despite its similar expression. We study the question of why the VNP-completeness proof of the permanent fails for the determinant. We isolate three fundamental properties that are sufficient to prove a polynomial sequence is VNP-hard, of which two are shared by both the permanent and the determinant. We proceed to show that the permanent satisfies the third property, which we refer to as the ``cost of a boolean sum," while the determinant does not, showcasing the fundamental difference between the polynomial families. We further note that this differentiation also applies in the border complexity setting and that our results apply for counting complexity.

cs.CC↗

Approximation Depth of Convex Polytopes

We study approximations of polytopes in the standard model for computing polytopes using Minkowski sums and (convex hulls of) unions. Specifically, we study the ability to approximate a target polytope by polytopes of a given depth. Our main results imply that simplices can only be ``trivially approximated''. On the way, we obtain a characterization of simplices as the only ``outer additive'' convex bodies.

math.MG↗

Fixed and Periodic Points of the Intersection Body Operator

The intersection body $IK$ of a star-body $K$ in $\mathbb{R}^n$ was introduced by E. Lutwak following the work of H. Busemann, and plays a central role in the dual Brunn-Minkowski theory. We show that when $n \geq 3$, $I^2 K = c K$ iff $K$ is a centered ellipsoid, and hence $I K = c K$ iff $K$ is a centered Euclidean ball, answering long-standing questions by Lutwak, Gardner, and Fish-Nazarov-Ryabogin-Zvavitch. To this end, we recast the iterated intersection body equation as an Euler-Lagrange equation for a certain volume functional under radial perturbations, derive new formulas for the volume of $I K$, and introduce a continuous version of Steiner symmetrization for Lipschitz star-bodies, which (surprisingly) yields a useful radial perturbation exactly when $n\geq 3$.

math.FA↗

Data Selection for ERMs

Learning theory has traditionally followed a model-centric approach, focusing on designing optimal algorithms for a fixed natural learning task (e.g., linear classification or regression). In this paper, we adopt a complementary data-centric perspective, whereby we fix a natural learning rule and focus on optimizing the training data. Specifically, we study the following question: given a learning rule $\mathcal{A}$ and a data selection budget $n$, how well can $\mathcal{A}$ perform when trained on at most $n$ data points selected from a population of $N$ points? We investigate when it is possible to select $n \ll N$ points and achieve performance comparable to training on the entire population. We address this question across a variety of empirical risk minimizers. Our results include optimal data-selection bounds for mean estimation, linear classification, and linear regression. Additionally, we establish two general results: a taxonomy of error rates in binary classification and in stochastic convex optimization. Finally, we propose several open questions and directions for future research.

cs.LG↗

Intuitive norms are Euclidean

We call a norm on $\mathbb{R}^n$ intuitive if for every points $p_1,\ldots,p_m$ in $\mathbb{R}^n$, one of the geometric medians of the points over the norm is in their convex hull. We characterize all intuitive norms.

math.MG↗

A Unified Characterization of Private Learnability via Graph Theory

We provide a unified framework for characterizing pure and approximate differentially private (DP) learnability. The framework uses the language of graph theory: for a concept class $\mathcal{H}$, we define the contradiction graph $G$ of $\mathcal{H}$. Its vertices are realizable datasets, and two datasets $S,S'$ are connected by an edge if they contradict each other (i.e., there is a point $x$ that is labeled differently in $S$ and $S'$). Our main finding is that the combinatorial structure of $G$ is deeply related to learning $\mathcal{H}$ under DP. Learning $\mathcal{H}$ under pure DP is captured by the fractional clique number of $G$. Learning $\mathcal{H}$ under approximate DP is captured by the clique number of $G$. Consequently, we identify graph-theoretic dimensions that characterize DP learnability: the clique dimension and fractional clique dimension. Along the way, we reveal properties of the contradiction graph which may be of independent interest. We also suggest several open questions and directions for future research.

cs.LG↗