This paper studies GAIL's global convergence for general MDP and nonlinear rewards.
problem Understanding when GAIL algorithms achieve global convergence for general MDP and nonlinear rewards.
method Characterization of global convergence for various policy gradient algorithms applied to GAIL.
result First systematic theoretical study of GAIL for global convergence.
New algorithm optimizes multi-objective outcomes in uncertain environments.
problem Optimizing global concave rewards in online Markov decision processes with multiple actions.
method No-regret algorithm based on online convex optimization and UCRL2, with a gradient threshold procedure.
result Non-stationary policy diversifies outcomes to optimize the global concave reward.
We study the global convergence of generative adversarial imitation learning for linear quadratic regulators, which is posed as minimax optimization. To address the challenges arising from non-convex-concave geometry, we analyze the alternating gradient algorithm and establish its Q-linear rate of convergence to a uniq…
New algorithm for fair ranking in contextual bandits with concave rewards.
problem Fair ranking in recommendation systems.
method Geometric interpretation of CBCR as optimization, Frank-Wolfe analyses.
result First algorithm with provably vanishing regret for CBCR.
Optimizes algorithms for non-concave bandit problems.
problem Optimizing algorithms for non-concave bandit problems.
method Unified zeroth-order optimization paradigm.
result Minimax-optimal algorithms in the dimension for low-rank generalized linear bandit problems.
Despite the success of single-agent reinforcement learning, multi-agent reinforcement learning (MARL) remains challenging due to complex interactions between agents. Motivated by decentralized applications such as sensor networks, swarm robotics, and power grids, we study policy evaluation in MARL, where agents with jo…
Least Squares EM converges globally for log-concave mixtures.
problem Location estimation in mixtures of two log-concave densities.
method Least Squares EM algorithm applied to log-concave mixtures.
result Least Squares EM converges globally to the true location parameter.
A new algorithm uses concavity in Gaussian processes to optimize decisions in bandit problems.
problem Optimizing decisions in sequential problems with context-dependent rewards.
method Proposes a UCB algorithm using a shape-constrained reward function estimator based on a Gaussian Process model with concavity constraints.
result Derives regret bounds for the proposed UCB algorithm.
The paper tackles rested bandits with non-decreasing and concave rewards, deriving lower bounds and an efficient algorithm.
problem Studying the sample complexity and optimal strategies for rested bandits with specific reward properties.
method Deriving regret lower bounds and designing an efficient algorithm R-ed-UCB with theoretical and empirical analysis.
result An efficient algorithm R-ed-UCB with a regret bound of O ~ ( T 2 3 ) \widetilde{\mathcal{O}}(T^{\frac{2}{3}}) O ( T 3 2 ) under certain conditions. We consider a contextual version of multi-armed bandit problem with global knapsack constraints. In each round, the outcome of pulling an arm is a scalar reward and a resource consumption vector, both dependent on the context, and the global knapsack constraints require the total consumption for each resource to be bel…
Algorithm tackles constrained reinforcement learning with concave-convex and knapsack constraints.
problem Constrained episodic reinforcement learning with concave rewards and convex constraints.
method Modular analysis with strong theoretical guarantees for concave-convex and knapsack settings.
result Significantly outperforms existing approaches in constrained episodic environments.
The paper provides tight bounds for improving multi-armed bandits problem.
problem Improving multi-armed bandits problem with concave reward functions.
method Upper and lower bounds for randomized online algorithms, providing an O ( k log k ) O(\sqrt{k} \log k) O ( k log k ) approximation. result Achieved nearly-tight approximation guarantees for the improving multi-armed bandits problem.
Concave regularization methods provide natural procedures for sparse recovery. However, they are difficult to analyze in the high dimensional setting. Only recently a few sparse recovery results have been established for some specific local solutions obtained via specialized numerical procedures. Still, the fundamental…
Proves a synthetic Lorentzian Cartan-Hadamard theorem.
problem Formulates and proves a theorem for Lorentzian geometry.
method Uses an appropriate notion of local concavity for Lorentzian (pre-)length spaces.
result Establishes existence and uniqueness of timelike geodesics.
Paper solves globally optimal k-means for low dimensional data.
problem Finding globally optimal k-means solutions for low dimensional data.
method Formulates as a concave assignment problem, iteratively solving small concave and large linear programming problems.
result Solves k-means to global optimality for large data sets with several clusters.
New algorithm for optimizing statistical utilities in bandits.
problem Optimizing statistical functionals of long-run reward distributions.
method Influence-function calculus for stochastic gradient estimation, entropic mirror-ascent algorithm.
result Regret bounds that separate optimization and estimation errors.
We consider the classical problem of sequential resource allocation where a decision maker must repeatedly divide a budget between several resources, each with diminishing returns. This can be recast as a specific stochastic optimization problem where the objective is to maximize the cumulative reward, or equivalently …
Two important goals of high-dimensional modeling are prediction and variable selection. In this article, we consider regularization with combined L 1 L_1 L 1 and concave penalties, and study the sampling properties of the global optimum of the suggested method in ultra-high dimensional settings. The L 1 L_1 L 1 -penalty provides th…
This work overcomes bias in concave multi-objective reinforcement learning.
problem Gradient bias in policy gradient methods for concave scalarized multi-objective reinforcement learning.
method Developed a Natural Policy Gradient (NPG) algorithm with a multi-level Monte Carlo (MLMC) estimator.
result Achieved optimal O ~ ( ε − 2 ) \widetilde{\mathcal{O}}(ε^{-2}) O ( ε − 2 ) sample complexity for computing an ε ε ε -optimal policy. Derives concavity inequality and estimates for k k k -Hessian equations.
problem Interior estimates and curvature estimates for k k k -Hessian equations. method Concavity inequality and semi-convexity condition.
result Interior estimates and Liouville-type result for semi-convex solutions.
New star-shaped acceptability indexes generalize existing methods.
problem Generalizing existing acceptability measures.
method Characterizing acceptability indexes through star-shaped risk measures and sets.
result Introducing concrete examples linked to various financial measures.
New IRL algorithm identifies optimal reward and policy from expert demonstrations.
problem Understanding reward functions from expert demonstrations with neural networks.
method Two-timescale single-loop IRL algorithm for neural network parameterized rewards.
result First IRL algorithm with non-asymptotic convergence guarantee and global optimality in neural network settings.
New algorithm solves optimization problems without submodularity.
problem Finding efficient solutions for optimization problems when submodularity does not hold.
method Parallel quasi-concave set optimization algorithm.
result Efficient globally optimal solution to maxi-min problems.
A new algorithm reduces regret in high-dimensional contextual bandits.
problem High-dimensional contextual bandits with unknown payoff functions.
method Developed an algorithm based on stochastic approximation for globally concave functions.
result Achieved regret $ ilde{O}(T^{ rac{d_x+1}{d_x+2}})$ for globally concave functions.
Introduces CSLC models to bridge deep generative models and classical algorithms.
problem Mode collapse and memorization issues in deep generative models and restrictive assumptions in classical algorithms.
method Introduces conditionally strongly log-concave (CSLC) models, factorizing data distribution into strongly log-concave conditional distributions.
result Efficient parameter estimation and sampling algorithms with theoretical guarantees for non-log-concave data distributions.
ICCNLS models complex relationships as convex and concave components.
problem Complex input-output relationships with affine ambiguity.
method Sub-gradient constrained affine functions, global orthogonality constraints, L1, L2, and elastic net regularisation.
result Improved predictive accuracy and model simplicity compared to conventional methods.
New algorithms sample from log concave distributions without gradient Lipschitz continuity.
problem Sampling from log concave distributions without gradient Lipschitz continuity.
method Two algorithms based on monotone polygonal (tamed) Euler schemes.
result Non-asymptotic 2-Wasserstein distance bounds between the process and target measure.
We analyze a reweighted version of the Kikuchi approximation for estimating the log partition function of a product distribution defined over a region graph. We establish sufficient conditions for the concavity of our reweighted objective function in terms of weight assignments in the Kikuchi expansion, and show that a…
Study properties of contact structures on symplectic disk bundles with concave boundaries.
problem Understanding the geometric properties of contact structures on concave boundaries of symplectic disk bundles.
method Use tools from toric geometry and algebraic torsion measurements from embedded contact homology.
result All such contact manifolds have a global contact toric structure, and can be tight or overtwisted.
Paper proposes new Langevin samplers for sampling from log-concave distributions with superlinear gradient growth.
problem Sampling from log-concave distributions with superlinear gradient growth.
method Proposes two novel discretizations of kinetic Langevin SDEs, showing contractivity and log-Sobolev inequality.
result Establishes non-asymptotic bounds in 2-Wasserstein distance between sampled distributions and target measures.
New algorithm improves GAIL for image sequences with global encoder and reward penalization.
problem Low-level, high-dimensional state input in GAIL framework.
method Global encoder and reward penalization mechanism.
result Significant performance improvement in low-level and high-dimensional tasks.
New method for RL with general utilities using variational policy gradient.
problem Optimizing policies with general concave utility functions in RL.
method Derives Variational Policy Gradient Theorem, develops variational Monte Carlo gradient estimation algorithm.
result Global convergence to optimal policy for general objectives, exponential convergence under strong convexity.
A generalized optimistic method for saddle point problems with improved complexity.
problem Solving convex-concave saddle point problems efficiently.
method Proposes a generalized optimistic method that includes the optimistic gradient method as a special case, handling constrained saddle point problems with composite objective functions and arbitrary norms.
result Best-known global iteration complexity bounds for first-, second-, and higher-order methods.
This paper tackles non-linear reward optimization in resource allocation problems.
problem Optimizing a non-linear function of long-term average rewards in resource allocation problems.
method Proposes model-based and model-free algorithms to learn optimal policies.
result Model-based algorithm achieves a regret of $\Tilde{O}\left(LKDS\sqrt{\frac{A}{T}}
ight)$ for K K K objectives combined with a concave L L L -Lipschitz function. This work analyzes how overparameterization aids GANs in reaching global saddle points.
problem Understanding the role of overparameterization in GANs for convergence to global saddle points.
method Theoretical and empirical analysis of overparameterized GANs with various architectures and datasets.
result GDA converges to a global saddle point in overparameterized GANs with certain assumptions.
Decentralized method solves saddle point problems with theoretical guarantees.
problem Solving saddle point problems in a decentralized network.
method Proximal point method adapted for decentralized networks.
result Converges to approximate stationarity with rate of O(1/√T).
New algorithms prove fast convergence in complex min-max problems.
problem Proving fast convergence in nonconvex min-max optimization.
method Hamiltonian Gradient Descent (HGD) and Consensus Optimization (CO) algorithms.
result HGD and CO achieve linear convergence in various settings.
Non-convex regularizers usually improve the performance of sparse estimation in practice. To prove this fact, we study the conditions of sparse estimations for the sharp concave regularizers which are a general family of non-convex regularizers including many existing regularizers. For the global solutions of the regul…
A network of spiking agents learns complex tasks using global reward signals.
problem Solving complex reinforcement learning tasks.
method A hierarchical network of GLM spiking agents, each modulating its firing policy based on local and global reward signals.
result A network of spiking agents can learn complex action representations to solve RL tasks.
RWR converges to global optimum in certain settings.
problem Proving convergence of RWR to optimal policy.
method Iterative learning with return-weighted log-likelihood.
result RWR converges to global optimum under certain conditions.
New bounds for SMC show its advantage over MCMC in multimodal distributions.
problem Estimating expectations under multimodal distributions with slow global mixing.
method Proves finite sample complexities for SMC with local mixing times, addressing bias through sequential resampling.
result SMC provides fully polynomial time approximation for multimodal problems.
Optimizes portfolios using CPT utility via convex optimization.
problem Maximizing CPT utility in portfolio selection.
method Minorization-maximization (MM) algorithm and convex-concave (CC) procedure.
result Problems can be solved globally and efficiently.
This paper tackles robust control of noisy systems with uncertain distributions.
problem Optimal control of sampled-data stochastic systems with multiplicative noise and distributional ambiguity.
method Develops a convex relaxation to handle the ``concave-max'' geometry and derives a probabilistic performance guarantee.
result Derives an explicit, non-asymptotic bound on the duality gap and proves robust viability conditions.
This paper improves understanding of GAIL's generalization and computational efficiency.
problem Understanding the theoretical properties of GAIL, especially its generalization and computational aspects.
method Investigates GAIL's theoretical properties, showing guarantees for generalization and computational efficiency.
result GAIL can be efficiently solved by stochastic first order optimization algorithms with sublinear convergence.
Strict concavity proven for growth indicator function of certain groups.
problem Proving strict concavity of growth indicator function for specific groups.
method Smoothness of Manhattan hypersurface and critical-exponent map.
result Strict concavity of growth indicator function for relatively Anosov groups.
This work finds mixed equilibria in machine learning problems using measures and simultaneous gradient ascent-descent.
problem Finding pure equilibria in machine learning problems is computationally hard.
method Entropic regularization, simultaneous gradient ascent-descent, and particle discretization in the Wasserstein metric.
result Global convergence towards the global equilibrium in mixed equilibria problems.
Paper introduces SGA for barycenter optimization in optimal transport.
problem Optimizing Wasserstein barycenter for discrete distributions.
method Sobolev gradient ascent algorithm tailored to Wasserstein geometry.
result SGA achieves convergence rate similar to subgradient descent.
Reinforcement learning for embodied agents is a challenging problem. The accumulated reward to be optimized is often a very rugged function, and gradient methods are impaired by many local optimizers. We demonstrate, in an experimental setting, that incorporating an intrinsic reward can smoothen the optimization landsc…