arXiv research

New algorithm adapts to changing action sets in sequential decision making.

RL agent learns to smoothly change lanes in a dynamic driving environment.

New action poisoning attacks improve LinUCB's performance by changing action signals.

New algorithm tracks changes in infinite action space rewards.

Let $ρ_0$ be an action of a Lie group on a manifold with boundary that is transitive on the interior. We study the set of actions that are topologically conjugate to $ρ_0$, up to smooth or analytic change of coordinates. We show that in many cases, including the compactifications of negatively curved symmetric spaces, …

We introduce a new online learning framework where, at each trial, the learner is required to select a subset of actions from a given known action set. Each action is associated with an energy value, a reward and a cost. The sum of the energies of the actions selected cannot exceed a given energy budget. The goal is to…

Algorithm optimizes bandit decisions with changing action sets using Gaussian processes.

Efficient algorithms for online learning with changing action sets, achieving no-approximate-regret guarantees.

New algorithm handles MDPs with unknown, changing rewards efficiently.

Causal Bayesian networks interpret actions as interventions to connect models to real-world outcomes.

Classifies special homogeneous curves with polynomial equations.

Inverse classification is the process of perturbing an instance in a meaningful way such that it is more likely to conform to a specific class. Historical methods that address such a problem are often framed to leverage only a single classifier, or specific set of classifiers. These works are often accompanied by naive…

AutoML struggles with climate change data, but offers potential improvements.

PRINCE provides interpretable explanations for recommender systems by removing minimal user actions.

This work introduces a method to attribute model performance drops to distribution shifts.

A successful response to climate change needs vast investments in low-carbon research, energy, and sustainable development. Governments can drive research, provide environmental regulation, and accelerate global development, but the necessary low-carbon investments of 2-3% GDP have yet to materialise. A new strategy to…

System tracks regulatory changes for compliance officers.

The paper develops a method to create robust control policies for robots using information bottlenecks.

We study Cohen-Macaulay actions, a class of torus actions on manifolds, possibly without fixed points, which generalizes and has analogous properties as equivariantly formal actions. Their equivariant cohomology algebras are computable in the sense that a Chang-Skjelbred Lemma, and its stronger version, the exactness o…

We propose algorithms for online principal component analysis (PCA) and variance minimization for adaptive settings. Previous literature has focused on upper bounding the static adversarial regret, whose comparator is the optimal fixed action in hindsight. However, static regret is not an appropriate metric when the un…

Safe imitation learning with a safety layer for flexible training.

Novel framework proves fast RL convergence in continuous spaces.

Study identifies change points in piecewise constant reward functions with fixed exploration budget.

Recourse explanations can become invalid if collective actions change statistical data.

Proposes a recourse algorithm for machine learning decisions.

Autonomous driving decision-making is a great challenge due to the complexity and uncertainty of the traffic environment. Combined with the rule-based constraints, a Deep Q-Network (DQN) based method is applied for autonomous driving lane change decision-making task in this study. Through the combination of high-level …

Linear contextual bandit is an important class of sequential decision making problems with a wide range of applications to recommender systems, online advertising, healthcare, and many other machine learning related tasks. While there is a lot of prior research, tight regret bounds of linear contextual bandit with infi…

New framework guides resource usage to achieve sublinear regret in adversarial settings.

Online learning algorithms are designed to perform in non-stationary environments, but generally there is no notion of a dynamic state to model constraints on current and future actions as a function of past actions. State-based models are common in stochastic control settings, but commonly used frameworks such as Mark…

Improved algorithm detects changes in RL environments with non-stationary MDPs.

Optimizes sensor usage for detecting abrupt changes in sensor data.

A group action is called polar if there exists an immersed submanifold (a section) which intersects all orbits orthogonally. Such group actions have been studied extensively on symmetric spaces. We show how to construct a manifold admitting a polar group action by prescribing their isotropy groups along a fundamental d…

Study on rapid policy changes in reinforcement learning.

The paper tackles energy management in buildings with PCM using dynamic programming.

We explore homotopies in quantum field theory formalism.

Proposes a topological model for partial equivariance in neural networks.

Study variations of Riemannian submersions to maintain geodesic fibers and positive curvatures.

We achieve a finite regret bound of O(dlogd) for online inverse linear optimization with M-convex action sets.

Proposes an efficient method for ordered counterfactual explanations.

Differential invariants of a (pseudo)group action can vary when restricted to invariant submanifolds (differential equations). The algebra is still governed by the Lie-Tresse theorem, but may change a lot. We describe in details the case of the motion group $O(n)\ltimes\R^n$ acting on the full (unconstraint) jet-space …

For large-scale industrial processes under closed-loop control, process dynamics directly resulting from control action are typical characteristics and may show different behaviors between real faults and normal changes of operating conditions. However, conventional distributed monitoring approaches do not consider the…

A new concept of causality for abstract phenomena.

The main result of this paper asserts that if a Seifert fibered 4-manifold has nonzero Seiberg-Witten invariant, the homotopy class of regular fibers has infinite order. This is a nontrivial obstruction to smooth circle actions; as applications, we show how to destroy smooth circle actions on a 4-manifold by knot surge…

KeRNS tackles non-stationary reinforcement learning in metric spaces.

Many policy gradient methods are variants of Actor-Critic (AC), where a value function (critic) is learned to facilitate updating the parameterized policy (actor). The update to the actor involves a log-likelihood update weighted by the action-values, with the addition of entropy regularization for soft variants. In th…

One important effect of price shocks in the United States has been increased political attention paid to the structure and performance of oil and natural gas markets, along with some governmental support for energy conservation. This paper describes how price changes helped lead the emergence of a political agenda acco…

Let $X$ be a compact connected strongly pseudoconvex CR manifold of dimension $2n+1, n\ge 1$ with a transversal CR $S^1$-action on $X$. In this paper we introduce the Quillen metric on the determinant line of the Fourier components of the Kohn-Rossi cohomology on $X$ with respect to the $S^1$-action. We study the behav…

It is well established that humans decision making and instrumental control uses multiple systems, some which use habitual action selection and some which require deliberate planning. Deliberate planning systems use predictions of action-outcomes using an internal model of the agent's environment, while habitual action…