arXiv research

New bandit algorithm works without realizability assumption.

New active learning framework for multiclass classification beyond realizability assumption.

Theory extends optimal learning rates without realizability assumption.

New offline RL method works with limited data and function approximators.

Paper tackles offline RL with weak assumptions on both function classes and data coverage.

New method efficiently evaluates policies using trajectory data.

Unified framework for realizable and agnostic learning.

This work provides guarantees for off-policy function estimation under realizability assumptions.

A group is properly 3-realizable if it is the fundamental group of a compact polyhedron whose universal covering is proper homotopically equivalent to some 3-manifold. We prove that when such a group is also quasi-simply filtered then it has {\em pro-(finitely generated free) fundamental group at infinity} and {\em sem…

TensorPlan algorithm finds δ-optimal policies with poly$(H,d)$ queries under linearly realizable state-value function.

Characterizes concept classes for optimistic online learning.

Faster algorithm reduces contextual bandit regret with fewer offline regression calls.

Solves a problem related to Nielsen realization for certain groups.

We show that a continuous local semiflow of $C^k$-maps on a finite-dimensional $C^k$-manifold M can be embedded into a local $C^k$-flow on M under some weak (necessary) assumptions. This result is applied to an open problem in [fil/tei:01]. We prove that finite-dimensional realizations for interest rate models are high…

We obtain a bound for the area of a capillary $H-$surface in a three-manifold with umbilic boundary and controlled sectional curvature. We then analyze the geometry when this area bound is realized, and obtain rigidity theorems. As a side product, we obtain existence of totally geodesic embedded surfaces in hyperbolic …

New method for distributional off-policy evaluation using Bellman residual minimization.

This paper tackles deferral learning with multiple experts, providing strong theoretical guarantees.

Value-function approximation methods that operate in batch mode have foundational importance to reinforcement learning (RL). Finite sample guarantees for these methods often crucially rely on two types of assumptions: (1) mild distribution shift, and (2) representation conditions that are stronger than realizability. H…

New algorithms estimate Q-functions under partial coverage and realizability, improving offline RL guarantees.

Procyclicality of historical risk measure estimation means that one tends to over-estimate future risk when present realized volatility is high and vice versa under-estimate future risk when the realized volatility is low. Out of it different questions arise, relevant for applications and theory: What are the factors w…

New algorithm trains ReLU gates provably in linear time.

The paper optimizes RV estimation by efficient sampling in time-changed diffusion models.

Breaks the hardness conjecture for batch RL with a novel tournament-based approach.

GEEN uses deep learning to estimate unobserved variables from observed data.

No-regret learning fails to converge to Nash equilibria in mixed strategies.

Cooperation information sharing is important to theories of human learning and has potential implications for machine learning. Prior work derived conditions for achieving optimal Cooperative Inference given strong, relatively restrictive assumptions. We relax these assumptions by demonstrating convergence for any disc…

We consider the multi-armed bandit setting with a twist. Rather than having just one decision maker deciding which arm to pull in each round, we have $n$ different decision makers (agents). In the simple stochastic setting, we show that a "free-riding" agent observing another "self-reliant" agent can achieve just $O(1)…

A major challenge in contextual bandits is to design general-purpose algorithms that are both practically useful and theoretically well-founded. We present a new technique that has the empirical and computational advantages of realizability-based approaches combined with the flexibility of agnostic methods. Our algorit…

The paper explores learning good policies from past data in large state spaces.

Trajectory data suffices for efficient RL in linear MDPs.

Study bounds noise level in linear regression with dependent data.

This paper proposes a new RV prediction model using neural distributional transformation and co-training.

New RL method learns to skip states in linearly $q^π$-realizable MDPs, simplifying to linear MDPs.

Kernel Bayes' rule has been proposed as a nonparametric kernel-based method to realize Bayesian inference in reproducing kernel Hilbert spaces. However, we demonstrate both theoretically and experimentally that the prediction result by kernel Bayes' rule is in some cases unnatural. We consider that this phenomenon is i…

Unified HS and related methods with explicit modeling assumptions.

Study shows offline RL with partial coverage and weak function classes is possible.

Improved private agnostic learning with near-optimal sample complexity.

A viable and still unproved conjecture states that, if $X$ is a smooth algebraic surface and $C$ is a smooth algebraic curve in $X$, then $C$ realizes the smallest possible genus amongst all smoothly embedded $2$-manifolds in its homology class. A proof is announced here for this conjecture, for a large class of surfac…

The paper analyzes the performance of empirical risk minimization for $p$-norm linear regression.

New method estimates VaR and ES using high-frequency data, outperforming existing approaches.

New approach uses 'forward-looking' counterfactuals for treatment choice.

New algorithm tackles non-linear utility in MNL bandits with $ilde{O}(\sqrt{T})$ regret.

The basic financial purpose of a firm is to maximize its value. An inventory management system should also contribute to realization of this basic aim. Many current asset management models currently found in financial management literature were constructed with the assumption of book profit maximization as basic aim. H…

This paper studies universal rates of ERM for binary classification under agnostic learning.

Efficient RL algorithm for MDPs with linear $Q^π$ realizability, achieving optimal regret bound.

We discuss the probabilistic properties of the variation based third and fourth moments of financial returns as estimators of the actual moments of the return distributions. The moment variations are defined under non-parametric assumptions with quadratic variation method but for the computational tractability, we use …

Study stabilizes Morse-Bott cohomology for equivariant manifolds.

Study on learning to predict dynamical systems without assuming their structure.