This paper studies replicability in machine learning tasks from a geometric viewpoint. Recent works have revealed the role of geometric partitions and Sperner's lemma (and its variations) in designing replicable learning algorithms and in establishing impossibility results. A partition $\mathcal{P}$ of $\mathbb{R}^d$ is called a $(k,\epsilon)$-secluded partition if for every $\vec{p}\in\mathbb{R}^d$, an $\varepsilon$-radius ball (with respect to the $\ell_{\infty}$ norm) centered at $\vec{p}$ intersects at most $k$ members of $\mathcal{P}$. In relation to replicable learning, the parameter $k$ is closely related to the $\textit{list complexity}$, and the parameter $\varepsilon$ is related to the sample complexity of the replicable learner. Construction of secluded partitions with better parameters (small $k$ and large $\varepsilon$) will lead to replicable learning algorithms with small list and sample complexities. Motivated by this connection, we undertake a comprehensive study of secluded partitions and establish near-optimal relationships between $k$ and $\varepsilon$. 1. We show that for any $(k,\epsilon)$-secluded partition where each member has at most unit measure, it must be that $k \geq(1+2\varepsilon)^d$, and consequently, for the interesting regime $k\in[2^d]$ it must be that $\epsilon\leq\frac{\log_4(k)}{d}$. 2. To complement this upper bound on $\epsilon$, we show that for each $d\in\mathbb{N}$ and each viable $k\in[2^d]$, a construction of a $(k,\epsilon)$-secluded (unit cube) partition with $\epsilon\geq\frac{\log_4(k)}{d}\cdot\frac{1}{8\log_4(d+1)}$. This establishes the optimality of $\epsilon$ within a logarithmic factor. 3. Finally, we adapt our proof techniques to obtain a new ``neighborhood'' variant of the cubical KKM lemma (or cubical Sperner's lemma): For any coloring of $[0,1]^d$ in which no color is used on opposing faces, it holds for each $\epsilon\in(0,\frac12]$ that there is a point where the open $\epsilon$-radius $\ell_\infty$-ball intersects at least $(1+\frac23\epsilon)^d$ colors. While the classical Sperner/KKM lemma guarantees the existence of a point that is "adjacent" to points with $(d+1)$ distinct colors, the neighborhood version guarantees the existence of a small neighborhood with exponentially many points with distinct colors.

We investigate replicable learning algorithms. Informally a learning algorithm is replicable if the algorithm outputs the same canonical hypothesis over multiple runs with high probability, even when different runs observe a different set of samples from the unknown data distribution. In general, such a strong notion of replicability is not achievable. Thus we consider two feasible notions of replicability called {\em list replicability} and {\em certificate replicability}. Intuitively, these notions capture the degree of (non) replicability. The goal is to design learning algorithms with optimal list and certificate complexities while minimizing the sample complexity. Our contributions are the following. 1. We first study the learning task of estimating the biases of $d$ coins, up to an additive error of $\varepsilon$, by observing samples. For this task, we design a $(d+1)$-list replicable algorithm. To complement this result, we establish that the list complexity is optimal, i.e there are no learning algorithms with a list size smaller than $d+1$ for this task. We also design learning algorithms with certificate complexity $\tilde{O}(\log d)$. The sample complexity of both these algorithms is $\tilde{O}(\frac{d^2}{\varepsilon^2})$ where $\varepsilon$ is the approximation error parameter (for a constant error probability). 2. In the PAC model, we show that any hypothesis class that is learnable with $d$-nonadaptive statistical queries can be learned via a $(d+1)$-list replicable algorithm and also via a $\tilde{O}(\log d)$-certificate replicable algorithm. The sample complexity of both these algorithms is $\tilde{O}(\frac{d^2}{\nu^2})$ where $\nu$ is the approximation error of the statistical query. We also show that for the concept class \dtep, the list complexity is exactly $d+1$ with respect to the uniform distribution. To establish our upper bound results we use rounding schemes induced by geometric partitions with certain properties. We use Sperner/KKM Lemma to establish the lower bound results.

A probabilistic algorithm A is pseudodeterministic if, on every input, there exists a canonical value that is output with high probability. If the algorithm outputs one of k canonical values with high probability, then it is called a k-pseudodeterministic algorithm. In the study of pseudodeterminism, the Acceptance Probability Estimation Problem (APEP), which is to additively approximate the acceptance probability of a Boolean circuit, is emerging as a central computational problem. This problem admits a 2-pseudodeterministic algorithm. Recently, it was shown that a pseudodeterministic algorithm for this problem would imply that any multi-valued function that admits a k-pseudodeterministic algorithm for a constant k (including approximation algorithms) also admits a pseudodeterministic algorithm (Dixon, Pavan, Vinodchandran; ITCS 2021).

We exhibit several computational problems that are complete for multi-pseudodeterministic computations in the following sense: (1) these problems admit 2-pseudodeterministic algorithms (2) if there exists a pseudodeterministic algorithm for any of these problems, then any multi-valued function that admits a k-pseudodeterministic algorithm for a constant k, also admits a pseudodeterministic algorithm. We also show that these computational problems are complete for Search-BPP: a pseudodeterministic algorithm for any of these problems implies a pseudodeterministic algorithm for all problems in Search-BPP.

We investigate the notion of pseudodeterminstic approximation algorithms. A randomized approximation algorithm A for a function f is pseudodeterministic if for every input x there is a unique value v so that A(x) outputs v with high probability, and v is a good approximation of f(x). We show that designing a pseudodeterministic version of Stockmeyer's well known approximation algorithm for the NP-membership counting problem will yield a new circuit lower bound: if such an approximation algorithm exists, then for every k, there is a language in the complexity class ZPP^{NP}_{tt} that does not have n^k-size circuits. While we do not know how to design such an algorithm for the NP-membership counting problem, we show a general result that any randomized approximation algorithm for a counting problem can be transformed to an approximation algorithm that has a constant number of influential random bits. That is, for most settings of these influential bits, the approximation algorithm will be pseudodeterministic.

Investigating the complexity of randomized space-bounded machines that are allowed to make multiple passes over the random tape has been of recent interest. In particular, it has been shown that derandomizing such probabilistic machines yields a weak but new derandomization of probabilistic time-bounded classes. In this paper we further explore the complexity of such machines. In particular, as our main result we show that for any epsilon<1, every language that is accepted by an O(n^epsilon)-pass, randomized logspace machine can be simulated in deterministic logspace with linear amount of advice. This result extends an earlier result of Fortnow and Klivans who showed that RL is in deterministic logspace with linear advice.