Paper establishes first instance-dependent lower bound for PAC reinforcement learning.
problem Identifying near-optimal policies in tabular MDPs with minimal samples.
method Proposes instance-dependent lower bound for sample complexity.
result Lower bound closely matches PEDEL algorithm's sample complexity.
Paper develops PAC verification for hypothesis classes and statistical algorithms.
problem Verifying machine learning models interactively.
method Develops interactive proof for PAC verification, proves lower bounds, and introduces a generalization.
result Improved protocol for verifying unions of intervals and statistical query algorithms.
Paper establishes new lower bounds for MDPs with changing transition kernels.
problem Minimizing sample complexity and regret in non-stationary MDPs.
method Developed novel lower bounds and constructed hard MDPs.
result Proved Ω ( ( H 3 S A / ε 2 ) log ( 1 / δ ) ) Ω((H^3SA/ε^2)\log(1/δ)) Ω (( H 3 S A / ε 2 ) log ( 1/ δ )) sample complexity lower bound. PAC-Bayesian bounds for MLPs with cross entropy loss validated.
problem Generalization bounds for MLPs with cross entropy loss.
method Introduced probabilistic explanations and proved PAC-Bayesian bounds using ELBO.
result MLPs with cross entropy loss inherently guarantee PAC-Bayesian generalization bounds.
This work establishes a new upper bound on the number of samples sufficient for PAC learning in the realizable case. The bound matches known lower bounds up to numerical constant factors. This solves a long-standing open problem on the sample complexity of PAC learning. The technique and analysis build on a recent brea…
Investigates tight PAC-Bayes bounds for small datasets.
problem Tightening PAC-Bayes bounds for small data.
method Generic PAC-Bayes theorem, meta-learning, synthetic tasks.
result PAC-Bayes bounds are competitive with Chernoff bounds but not as tight.
This paper explores the adaptive (active) PAC (probably approximately correct) top- k k k ranking (i.e., top- k k k item selection) and total ranking problems from l l l -wise ( l ≥ 2 l\geq 2 l ≥ 2 ) comparisons under the multinomial logit (MNL) model. By adaptively choosing sets to query and observing the noisy output of the most favored …
Proves SQ lower bounds for learning two-hidden-layer neural networks.
problem Learning two-hidden-layer ReLU networks with Gaussian inputs.
method Refined lifting procedure to reduce Boolean PAC learning to Gaussian.
result Superpolynomial SQ lower bounds for Gaussian inputs.
New algorithm for learning ReLU networks with Gaussian noise, improving previous results.
problem PAC learning one-hidden-layer ReLU networks with Gaussian marginals and label noise.
method First polynomial-time algorithm for k k k up to i l d e O ( log d ) ilde{O}(\sqrt{\log d}) i l d e O ( log d ) for positive coefficients, no assumptions on rank or condition number. result Proves a Statistical Query lower bound of d Ω ( k ) d^{Ω(k)} d Ω ( k ) for arbitrary real coefficients, separating learnability classes. Paper bounds PAC RL sample complexity in deterministic MDPs.
problem Identify ε-optimal policy with high probability.
method Proposes nearly matching upper and lower bounds on sample complexity, introduces deterministic return gap, uses graph-theoretical concepts and maximum-coverage exploration.
result First nearly matching upper and lower bounds on sample complexity for PAC RL in deterministic MDPs.
Deep neural networks are optimal for dependent data using PAC-Bayes bounds.
problem Optimizing deep neural networks for dependent data.
method PAC-Bayes oracle inequalities and Bernstein inequality.
result Upper and lower bounds match, proving minimax optimality.
New bounds on learning from multiple distributions for VC classes.
problem Understanding the sample complexity of learning from multiple data distributions.
method Analyzing the gap between known upper and lower bounds for PAC-learnable classes.
result Recent progress on sample complexity for VC dimension d classes on k distributions.
New combinatorial dimension VCL refines learning curve theory.
problem Explaining the behavior of learning curves for specific distributions.
method Introducing combinatorial dimension VCL to characterize strong minimax lower bounds.
result Learning rate can be decomposed into linear and exponential components.
Private RL algorithm with privacy guarantees for personalized medicine decisions.
problem Privacy-preserving reinforcement learning for personalized medicine decisions.
method Developed a private optimism-based RL algorithm using joint differential privacy (JDP).
result Achieved strong PAC and regret bounds with a privacy guarantee.
This paper establishes strong lower bounds for learning in revealing POMDPs.
problem Understanding the fundamental limits of reinforcement learning in revealing partially observable Markov Decision Processes (POMDPs).
method Develops strong PAC and regret lower bounds for learning in revealing POMDPs using multi-step revealing POMDPs as a case study.
result Strong polynomial lower bounds for learning in revealing POMDPs, achieving significantly smaller gaps against current upper bounds.
New method improves training stochastic neural networks with tighter guarantees.
problem Training stochastic neural networks with provable guarantees.
method Developed partially-aggregated estimators and reformulated PAC-Bayesian bounds.
result Derives a differentiable objective leading to tighter generalisation guarantees.
New algorithm for reliable learning of Gaussian halfspaces with improved sample and computational complexity.
problem Learning halfspaces under Gaussian marginals with reliable agnostic model.
method Developed a new algorithm for reliable learning of Gaussian halfspaces with specific sample and computational complexity.
result Achieved a new algorithm with improved sample and computational complexity for reliable learning of Gaussian halfspaces.
Recently, there has been significant progress in understanding reinforcement learning in discounted infinite-horizon Markov decision processes (MDPs) by deriving tight sample complexity bounds. However, in many real-world applications, an interactive learning agent operates for a fixed or bounded period of time, for ex…
PAC-Bayesian analysis improves lifelong learning in multi-armed bandits.
problem Improving lifelong learning in multi-armed bandits.
method PAC-Bayesian analysis for deriving lower bounds and proposing lifelong learning algorithms.
result Proposed algorithms outperform baseline methods in lifelong multi-armed bandit problems.
Modeling alignment as resource-limited cognitive processes, researchers derive performance bounds.
problem Systematic deviations in feedback-based alignment of large language models.
method Modeling alignment as a two-stage cascade U o H o Y U o H o Y U oH o Y given S S S , with cognitive and total capacities. result Capacity-coupled Alignment Performance Interval derived from Fano and PAC-Bayes bounds.
PROWL uses robust reward estimates to improve ITR selection.
problem Reward uncertainty in ITR estimation leads to inflated performance.
method PAC-Bayesian framework with reward uncertainty certificates.
result PROWL achieves better robust treatment regime estimation.
New framework improves model reliability under distribution shifts.
problem Lack of formal guarantees connecting shift magnitude to prediction reliability in TTA methods.
method Develops a PAC-Bayesian framework interpreting MMD-balls as credal sets.
result Establishes generalization bounds and provides epistemic uncertainty quantification.
We consider the problem of identifying any k k k out of the best m m m arms in an n n n -armed stochastic multi-armed bandit. Framed in the PAC setting, this particular problem generalises both the problem of `best subset selection' and that of selecting `one out of the best m' arms [arcsk 2017]. In applications such as crowd…
Study risk-sensitive reinforcement learning with optimized certainty equivalents.
problem Risk-sensitive reinforcement learning in finite discounted MDPs.
method Analyzed a simple model-based approach and derived PAC sample complexity bounds.
result Established tight sample complexity bounds for value and policy learning.
Learning sentence vectors from an unlabeled corpus has attracted attention because such vectors can represent sentences in a lower dimensional and continuous space. Simple heuristics using pre-trained word vectors are widely applied to machine learning tasks. However, they are not well understood from a theoretical per…
We consider the Max K K K -Armed Bandit problem, where a learning agent is faced with several stochastic arms, each a source of i.i.d. rewards of unknown distribution. At each time step the agent chooses an arm, and observes the reward of the obtained sample. Each sample is considered here as a separate item with the rewa…
This paper optimizes the number of comparisons needed to find the best k items from pairwise comparisons.
problem Finding the best k items from pairwise comparisons with limited comparisons.
method Developed algorithms for finding probably approximately correct and exact best k items under stochastic conditions.
result Upper and lower bounds on the number of comparisons for finding the best k items, with matching upper bounds for PAC best k items.
Exact minimax risk derived for linear prediction with sample covariance analysis.
problem Understanding the minimax risk in linear prediction under various covariate distributions.
method Exact minimax risk analysis, leveraging statistical leverage scores and PAC-Bayes techniques.
result The minimax risk is of order d / ( n − d + 1 ) d/(n-d+1) d / ( n − d + 1 ) for any covariate distribution, nearly matching the risk for Gaussian design. We consider the Max K K K -Armed Bandit problem, where a learning agent is faced with several sources (arms) of items (rewards), and interested in finding the best item overall. At each time step the agent chooses an arm, and obtains a random real valued reward. The rewards of each arm are assumed to be i.i.d., with an un…
PAC-MCTS addresses biased search in LLM-guided planning by dynamically pruning.
problem Systematic biases in LLMs lead to inefficient and unsafe search in deep planning tasks.
method Formulates node expansion as BAI under bounded bias, derives sample complexity bounds, and proposes PAC-MCTS for dynamic confidence bounds.
result PAC-MCTS improves robustness and efficiency by up to 78% fewer API evaluations and 3x higher sample efficiency.
Study shows SQ lower bounds for learning ReLUs with Massart noise.
problem Learning a single neuron in the presence of Massart noise.
method Statistical Query (SQ) lower bounds for efficient learning algorithms.
result No efficient SQ algorithm can approximate the optimal error within any constant factor.
This research improves PAC-Bayesian bounds for classification tasks using convexified loss.
problem Deriving generalization bounds for classification tasks with non-convex loss functions.
method Shift focus to misclassification excess risk bounds for PAC-Bayesian classification using convex surrogate loss and leveraging PAC-Bayesian relative bounds in expectation.
result Improved PAC-Bayesian bounds for classification tasks with convex surrogate loss.
Compact learning results across various loss functions.
problem Understanding sample complexity in transductive learning.
method Analyzing finite projections and sample complexities for different loss functions.
result Exact compactness of sample complexity holds broadly across realizable and agnostic learning.
Meta-learning bounds derived using PAC-Bayes theory for improved generalization.
problem Uncertainty in generalization performance for meta-learning with new tasks.
method PAC-Bayes relative entropy bounds and empirical risk minimization (ERM) method.
result Competitive generalization performance and rapid convergence with data-dependent prior.
Unified derivation of PAC-Bayes and MI bounds for general VC classes with fast rates.
problem Generalization bounds for machine learning models with VC classes.
method Unified derivation of conditional PAC-Bayesian and mutual information bounds, including MAC-Bayesian bounds.
result Nontrivial bounds for general VC classes and faster rates for specific conditions.
Study shows exponential gap in sample complexity between noisy and non-noisy recurrent neural networks.
problem Understanding the impact of noise on the sample complexity of recurrent neural networks.
method Analyzing noisy multi-layered sigmoid recurrent neural networks with independent noise and proving lower bounds.
result Exponential gap in sample complexity between noisy and non-noisy networks, even for small noise values.
The paper aims to develop new combinatorial dimensions for bounded memory learning.
problem Characterize bounded memory learning using combinatorial dimensions.
method Proposes a candidate solution based on the SQ dimension of neighboring distributions and proves upper and lower bounds.
result Characterizes bounded memory learning in a specific parameter regime, matching equivalence between bounded memory and SQ learning.
Study shows multi-source learning is more resilient to adversarial corruption than single-source learning.
problem Learning from multiple untrusted data sources, especially when some are adversarially corrupted.
method Analyzed the scenario where an adversary can corrupt a fixed fraction of data sources, derived a generalization bound for this setting.
result PAC-learnability is possible in the multi-source setting even when some data sources are adversarially corrupted.
New algorithm optimizes PAC-Bayes bound without surrogate loss.
problem Mismatch between optimisation objective and generalisation bound in stochastic neural networks.
method Proposes a novel training algorithm that optimizes the PAC-Bayesian bound directly.
result Empirical results show improved performance over existing PAC-Bayesian training methods.
This paper improves meta-learning by developing new PAC-Bayes bounds.
problem Meta-learning generalization gap across multiple tasks.
method Upper bounding convex functions linking environment and task-level losses.
result New PAC-Bayes bounds for meta-learning with improved algorithms.
Improved mistake bounds for transductive online learning.
problem Quantifying the power of unlabeled data in online learning.
method Proving lower and upper bounds on transductive mistake bounds.
result Exponential improvement in mistake bounds for transductive learning.
We consider the problems of robust PAC learning from distributed and streaming data, which may contain malicious errors and outliers, and analyze their fundamental complexity questions. In particular, we establish lower bounds on the communication complexity for distributed robust learning performed on multiple machine…
New algorithm reduces contextual bandit identification to argmax calls.
problem Best-arm identification in stochastic contextual bandits.
method Instance-optimal PAC algorithm using argmax oracle calls.
result First instance-dependent PAC sample complexity for contextual bandits.
We consider PAC-learning a good item from k k k -subsetwise feedback information sampled from a Plackett-Luce probability model, with instance-dependent sample complexity performance. In the setting where subsets of a fixed size can be tested and top-ranked feedback is made available to the learner, we give an algorithm w…
New empirical PAC-Bayes bound for Markov chains with finite state space.
problem Lack of empirical bounds for Markov chains with temporal dependence.
method Proved a new PAC-Bayes bound for Markov chains, providing an empirical pseudo-spectral gap.
result First fully empirical PAC-Bayes bound for Markov chains with finite state space.
Survey and compare PAC-Bayes bounds for bandit problems.
problem Designing and evaluating bandit algorithms with strong performance guarantees.
method PAC-Bayes bounds applied to bandit problems.
result PAC-Bayes bounds useful for offline bandit algorithms, but loose for online algorithms.
Introduces PAC-Bayes bounds for understanding learning procedures.
problem Understanding the generalization ability of learning procedures.
method PAC-Bayesian bounds and their applications to neural networks.
result Simplified version of localization technique described.
PAC-Bayes framework fails on simple 1D linear classification task.
problem Proving the learnability of simple 1D linear classification tasks using PAC-Bayes bounds.
method Demonstrated a specific 1D linear classification task that PAC-Bayes cannot analyze.
result PAC-Bayes framework cannot prove learnability of simple 1D linear classification tasks.