AGAN is the first neural architecture search method for GANs that discovers architectures outperforming state-of-the-art on CIFAR-10 unsupervised image generation and competitive on supervised tasks.
hub Canonical reference
Neural Combinatorial Optimization with Reinforcement Learning
Canonical reference. 100% of citing Pith papers cite this work as background.
46
Pith papers citing it
278
external citations · Pith
Background
100% of classified citations
abstract
This paper presents a framework to tackle combinatorial optimization problems using neural networks and reinforcement learning. We focus on the traveling salesman problem (TSP) and train a recurrent network that, given a set of city coordinates, predicts a distribution over different city permutations. Using negative tour length as the reward signal, we optimize the parameters of the recurrent network using a policy gradient method. We compare learning the network parameters on a set of training graphs against learning them on individual test graphs. Despite the computational expense, without much engineering and heuristic designing, Neural Combinatorial Optimization achieves close to optimal results on 2D Euclidean graphs with up to 100 nodes. Applied to the KnapSack, another NP-hard problem, the same method obtains optimal solutions for instances with up to 200 items.
hub tools
citation-role summary
background 5
citation-polarity summary
fields
cs.LG 22 cs.AI 8 quant-ph 4 cs.CL 2 cs.CV 2 math.OC 2 cond-mat.dis-nn 1 cs.IR 1 cs.MA 1 cs.RO 1roles
background 5polarities
background 5representative citing papers
A reinforcement learning policy for the vertex-guard art gallery problem encodes sufficient geometric information in its encoder to allow a simple classifier to achieve high coverage feasibility out of distribution.
EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.
TriSearch is an RL framework that optimizes triangulations of polytopes using bistellar flips with a circuit-supported subtriangulation action representation, generalizing zero-shot to larger instances and outperforming prior samplers in 3D and 4D.
SPACE framework unifies symmetric and asymmetric VRPs via bidirectional Frechet representations and weight-decomposed decoding for zero-shot generalization across 110 variants.
MEMOIR adds branch-local and global memory with a reflection step to tree search for LLM solver synthesis, reaching 96.7% solution validity and 7.3-point score gains over baselines on seven CO problems with lower run-to-run variance.
TTT-Discover applies test-time RL to set new state-of-the-art results on math inequalities, GPU kernels, algorithm contests, and single-cell denoising using an open model and public code.
Dreamer 4 is the first agent to obtain diamonds in Minecraft from only offline data by reinforcement learning inside a scalable world model that accurately predicts game mechanics.
PLMA combines cross-graph attention EBMs with short warm-started MCMC chains to reach near-zero average optimality gaps on QAPLIB and strong robustness on hard Taixxeyy instances.
VaP-CSMV uses a cross-semantic encoder and multi-view decoder to unify DRL solving of HFVRP variants, outperforming prior neural solvers while matching heuristics at much lower inference time and generalizing zero-shot to unseen scales.
NCP trains a neural network to predict certificate-level dual prices for CO problems, enabling structured primal recovery with a local second-order error guarantee when consistency holds.
AtomTreeSearch embeds a neutral-atom quantum MWIS subroutine inside Monte Carlo Tree Search and matches or exceeds OR-Tools and simulated annealing on TSP instances up to 100 cities.
A geometry-enhanced non-autoregressive neural TSP solver with multi-candidate RL cuts the TSPLIB optimality gap from 17.12% to 3.60% while solving in milliseconds per instance.
RACL lets a reasoning agent discover and apply control rules to a metaheuristic by observing operational memory and testing bounded interventions, shown on vehicle routing with reported cost improvements over baselines.
MViewRouter internalizes D4 geometric equivariance for routing via Multi-view Alternating Attention and Collective Policy Gradient Aggregation, yielding competitive solutions and strong generalization on TSP/CVRP benchmarks.
PARCEL is a new visual tokenization architecture combining pool-anchored resampling with conditioned elastic queries to enhance performance-efficiency tradeoffs in LVLMs over prior matryoshka methods.
AlphaTransit pairs MCTS with a learned policy-value network to reach 54.6% and 82.1% service rates on a Bloomington transit benchmark, outperforming plain RL and plain MCTS baselines.
Deep Boltzmann Quantum States with natural-gradient optimization and annealing-like training match exact or best-known solutions for large infinite-range Ising spin glasses and solve job shop scheduling instances.
SCOPE-BENCH shows state-of-the-art molecular models suffer up to 8x higher errors under extreme OOD, while POMA reduces mean absolute error by up to 11.2% via target-aware source selection and dual-scale adaptation.
Reinforcement learning policy for qubit mapping reduces SWAP overhead by 65-85% versus standard quantum compilers on MQTBench and Queko benchmark circuits.
ECO combines a Mamba encoder-decoder with two-stage batched DPO and LS-aware preference construction, claiming the best reported neural performance and near-linear memory scaling on TSP up to 5000 nodes and CVRP up to 1000 nodes.
A differentiable MPNN approximates uniform facility location with provable guarantees and outperforms standard approximation algorithms while closing the gap to exact ILP solutions.
QARMA applies transformer-augmented reinforcement learning to qubit allocation and reuse in modular quantum systems, reporting up to 86% average reduction in inter-core communications versus optimized Qiskit baselines.
An attention-based DRL agent with Transformer encoder and GNN learns heuristics for qubit-to-core allocation in multi-core quantum systems to minimize state transfers and online compilation time.
citing papers explorer
-
AGAN: Towards Automated Design of Generative Adversarial Networks
AGAN is the first neural architecture search method for GANs that discovers architectures outperforming state-of-the-art on CIFAR-10 unsupervised image generation and competitive on supervised tasks.
-
Learning to Place Guards by Reinforcement: A Geo-Free Neural Policy for the Vertex-Guard Art Gallery Problem
A reinforcement learning policy for the vertex-guard art gallery problem encodes sufficient geometric information in its encoder to allow a simple classifier to achieve high coverage feasibility out of distribution.
-
Interpreting Neural Combinatorial Optimization via Evolving Programmatic Bottlenecks
EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.
-
TriSearch: Learning to Optimize Triangulations via Bistellar Flips
TriSearch is an RL framework that optimizes triangulations of polytopes using bistellar flips with a circuit-supported subtriangulation action representation, generalizing zero-shot to larger instances and outperforming prior samplers in 3D and 4D.
-
SPACE: Unifying Symmetric and Asymmetric Routing Problems for Generalist Neural Solver
SPACE framework unifies symmetric and asymmetric VRPs via bidirectional Frechet representations and weight-decomposed decoding for zero-shot generalization across 110 variants.
-
Memory-Guided Tree Search with Cross-Branch Knowledge Transfer for LLM Solver Synthesis
MEMOIR adds branch-local and global memory with a reflection step to tree search for LLM solver synthesis, reaching 96.7% solution validity and 7.3-point score gains over baselines on seven CO problems with lower run-to-run variance.
-
Learning to Discover at Test Time
TTT-Discover applies test-time RL to set new state-of-the-art results on math inequalities, GPU kernels, algorithm contests, and single-cell denoising using an open model and public code.
-
Training Agents Inside of Scalable World Models
Dreamer 4 is the first agent to obtain diamonds in Minecraft from only offline data by reinforcement learning inside a scalable world model that accurately predicts game mechanics.
-
Learning to Solve the Quadratic Assignment Problem with Warm-Started MCMC Finetuning
PLMA combines cross-graph attention EBMs with short warm-started MCMC chains to reach near-zero average optimality gaps on QAPLIB and strong robustness on hard Taixxeyy instances.
-
Vehicle-as-Prompt: A Unified Deep Reinforcement Learning Framework for Heterogeneous Fleet Vehicle Routing Problem
VaP-CSMV uses a cross-semantic encoder and multi-view decoder to unify DRL solving of HFVRP variants, outperforming prior neural solvers while matching heuristics at much lower inference time and generalizing zero-shot to unseen scales.
-
Neural Certificate Pricing for Combinatorial Optimization Problems
NCP trains a neural network to predict certificate-level dual prices for CO problems, enabling structured primal recovery with a local second-order error guarantee when consistency holds.
-
Quantum-enhanced Monte Carlo Tree Search framework for combinatorial optimization problems
AtomTreeSearch embeds a neutral-atom quantum MWIS subroutine inside Monte Carlo Tree Search and matches or exceeds OR-Tools and simulated annealing on TSP instances up to 100 cities.
-
GeoRouteNet: A Geometry-Aware Non-Autoregressive Neural Solver for the Euclidean Traveling Salesman Problem
A geometry-enhanced non-autoregressive neural TSP solver with multi-candidate RL cuts the TSPLIB optimality gap from 17.12% to 3.60% while solving in milliseconds per instance.
-
RACL: Reasoning-Agent Control Layers for Continuous Metaheuristic Learning
RACL lets a reasoning agent discover and apply control rules to a metaheuristic by observing operational memory and testing bounded interventions, shown on vehicle routing with reported cost improvements over baselines.
-
MViewRouter: Internalizing Geometric Equivariance via Multi-view Alternating Attention for Combinatorial Routing
MViewRouter internalizes D4 geometric equivariance for routing via Multi-view Alternating Attention and Collective Policy Gradient Aggregation, yielding competitive solutions and strong generalization on TSP/CVRP benchmarks.
-
PARCEL: Pool-Anchored Resampling with Conditioned Elastic Queries for Efficient Vision-Language Understanding
PARCEL is a new visual tokenization architecture combining pool-anchored resampling with conditioned elastic queries to enhance performance-efficiency tradeoffs in LVLMs over prior matryoshka methods.
-
AlphaTransit: Learning to Design City-scale Transit Routes
AlphaTransit pairs MCTS with a learned policy-value network to reach 54.6% and 82.1% service rates on a Bloomington transit benchmark, outperforming plain RL and plain MCTS baselines.
-
Solving Classical and Quantum Spin Glasses with Deep Boltzmann Quantum States
Deep Boltzmann Quantum States with natural-gradient optimization and annealing-like training match exact or best-known solutions for large infinite-range Ising spin glasses and solve job shop scheduling instances.
-
Rethinking Molecular OOD Generalization via Target-Aware Source Selection
SCOPE-BENCH shows state-of-the-art molecular models suffer up to 8x higher errors under extreme OOD, while POMA reduces mean absolute error by up to 11.2% via target-aware source selection and dual-scale adaptation.
-
CO-MAP: A Reinforcement Learning Approach to the Qubit Allocation Problem
Reinforcement learning policy for qubit mapping reduces SWAP overhead by 65-85% versus standard quantum compilers on MQTBench and Queko benchmark circuits.
-
Rethinking Efficiency in Neural Combinatorial Optimization: Batched Preference Optimization with Mamba
ECO combines a Mamba encoder-decoder with two-stage batched DPO and LS-aware preference construction, claiming the best reported neural performance and near-linear memory scaling on TSP up to 5000 nodes and CVRP up to 1000 nodes.
-
Learning to Approximate Uniform Facility Location via Graph Neural Networks
A differentiable MPNN approximates uniform facility location with provable guarantees and outperforms standard approximation algorithms while closing the gap to exact ILP solutions.
-
Learning-Optimized Qubit Mapping and Reuse to Minimize Inter-Core Communication in Modular Quantum Architectures
QARMA applies transformer-augmented reinforcement learning to qubit allocation and reuse in modular quantum systems, reporting up to 86% average reduction in inter-core communications versus optimized Qiskit baselines.
-
Attention-Based Deep Reinforcement Learning for Qubit Allocation in Modular Quantum Architectures
An attention-based DRL agent with Transformer encoder and GNN learns heuristics for qubit-to-core allocation in multi-core quantum systems to minimize state transfers and online compilation time.
-
Contextual Plackett-Luce: An Efficient Neural Model for Probabilistic Sequence Selection under Ambiguity
Contextual Plackett-Luce extends the classical Plackett-Luce model with context-dependent Ising parameterization to enable efficient parallel scoring followed by incremental autoregressive selection for ambiguous sequence tasks.
-
HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization
HMACE deploys Proposer, Generator, Evaluator, and Reflector agents in an evolutionary loop to generate and refine heuristics for NP-hard problems, reporting lower optimality gaps and token costs than baselines on TSP and Online BPP.
-
Graph Normalization: Fast Binarizing Dynamics for Differentiable MWIS
Graph Normalization is a convergent dynamical system that approximates MWIS by always reaching a binary maximum independent set via majorization-minimization and evolutionary game equivalence.
-
Linear Decision Tree Policies for Integer Linear Programs
Any fixed integer linear program with a finite feasible set can be answered by a precomputed linear decision tree using polynomially many arithmetic operations per cost query; a practical construction works on small instances.
-
A Hybrid Reinforcement and Self-Supervised Learning Aided Benders Decomposition Algorithm
A hybrid RL and self-supervised learning method accelerates generalized Benders decomposition by 57.5% on a MINLP case study while recovering optimal solutions.
-
Neural Global Optimization via Iterative Refinement from Noisy Samples
A neural network with iterative refinement and spline-coefficient inputs reduces mean global-minimum error from 36.24% to 8.05% on 1D noisy B-spline functions.
-
Hierarchical Reinforcement Learning for Neural Network Compression (HiReLC): Pruning and Quantization
HiReLC applies hierarchical ensemble RL with surrogate guidance to jointly prune and quantize neural networks, reporting 5.99-6.72x compression with small accuracy changes on ViTs and CNNs.
-
N(CO)$^2$: Neural Combinatorial Optimization with Chance Constraints to Solve Stochastic Orienteering
N(CO)^2 applies reinforcement learning with chance constraints to solve stochastic orienteering problems, generalizing across instances with performance competitive to MILP.
-
Geometry-Aware Reinforcement Learning for 2D Irregular Nesting
Reinforcement learning paired with a geometry-aware Polygons Transformer achieves area utilization competitive with the Sparrow heuristic solver for 2D irregular nesting.
-
Smart Transportation Without Neurons -- Fair Metro Network Expansion with Tabular Reinforcement Learning
Tabular RL on a Non-Markovian Rewards Decision Process formulation matches deep RL performance on real metro expansion in Xi'an and Amsterdam while cutting episodes by 18x and carbon emissions by 12x on average.
-
Convex Compositional Reasoning Models
CCEM parameterizes compositional energy factors with input-convex neural networks and optimizes over a convex relaxation to enable deterministic scaling from small to large combinatorial reasoning instances.
-
ReVEL: Multi-Turn Reflective LLM-Guided Heuristic Evolution via Structured Performance Feedback
Grouping related heuristics and refining them jointly with multi-turn LLM feedback improves LLM-guided evolutionary search on combinatorial optimization benchmarks.
-
Combinatorial Keyword Recommendations for Sponsored Search with Deep Reinforcement Learning
A modified pointer network trained with actor-critic DRL and Equal Size K-Means clustering is applied to combinatorial keyword recommendation in sponsored search, reporting offline and online gains.
-
ARMATA: Auto-Regressive Multi-Agent Task Assignment
ARMATA is a new end-to-end autoregressive model with multi-stage decoding that unifies allocation and routing for multi-agent systems and reports up to 20% better solutions than OR-Tools, CPLEX, and LKH-3 in seconds instead of hours.
-
Machine Learning-based Two-Stage Graph Sparsification for the Travelling Salesman Problem
A two-stage ML pipeline unions α-Nearest and POPMUSIC candidate edges then prunes single-source edges via a classifier, cutting TSP graph density 37-47% with ≥99.69% optimal-tour recall.
-
Optimizing Nursing Care Taxi Dispatch Leveraging Integer Linear Programming Solvers and Machine Learning
A Transformer model trained via supervised learning on ILP solutions for a new nursing care taxi dispatch VRP variant reduces operating time by up to 8% on small instances while keeping constraint violations low.
-
PaliGemma 2: A Family of Versatile VLMs for Transfer
PaliGemma 2 is a family of vision-language models that achieves state-of-the-art results on transfer tasks like table structure recognition and radiography report generation by combining SigLIP with Gemma 2 models at various sizes and resolutions.
-
RL-SPH: Learning to Achieve Feasible Solutions for Integer Linear Programs
RL-SPH is an RL-based start primal heuristic that learns to turn infeasible starting points into feasible integer linear programming solutions, including problems with non-binary integers.
-
Finite Expression Method with TranNet-based Function Learning for High-Dimensional Partial Differential Equations
An extension of the finite expression method using TranNet-initialized shallow neural operators is proposed as an effective solver for high-dimensional partial differential equations.
-
Built Environment Reasoning from Remote Sensing Imagery Using Large Vision--Language Models
Large vision-language models applied to multi-scale remote sensing imagery can generate recommendations on built environment design, constructability, land use, and risks for smart city decision-making.
-
Gemma 2: Improving Open Language Models at a Practical Size
Gemma 2 models achieve leading performance at their sizes by combining established Transformer modifications with knowledge distillation for the 2B and 9B variants.
-
Deep Learning for Sequential Decision Making under Uncertainty: Foundations, Frameworks, and Frontiers
A tutorial framing deep learning as a complement to optimization for sequential decision-making under uncertainty, with applications in supply chains, healthcare, and energy.