Mechanistic Interpretability for Neural TSP Solvers

• URL: http://arxiv.org/abs/2510.21693v1
• Date: Fri, 24 Oct 2025 17:54:19 GMT
• Title: Mechanistic Interpretability for Neural TSP Solvers
• Authors: Reuben Narad, Leonard Boussioux, Michael Wagner,
• Abstract summary: We train a pointer network with reinforcement learning on 100-node instances, then fit an SAE to the encoder's residual stream to discover an overcomplete dictionary of interpretable features.<n>Our analysis reveals that the solver naturally develops features mirroring fundamental TSP concepts.<n>These findings provide the first model-internal account of what neural solvers compute before node selection, demonstrate that geometric structure emerges without explicit supervision, and suggest pathways toward transparent hybrid systems.
• Score: 0.8092772265574576
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Neural networks have advanced combinatorial optimization, with Transformer-based solvers achieving near-optimal solutions on the Traveling Salesman Problem (TSP) in milliseconds. However, these models operate as black boxes, providing no insight into the geometric patterns they learn or the heuristics they employ during tour construction. We address this opacity by applying sparse autoencoders (SAEs), a mechanistic interpretability technique, to a Transformer-based TSP solver, representing the first application of activation-based interpretability methods to operations research models. We train a pointer network with reinforcement learning on 100-node instances, then fit an SAE to the encoder's residual stream to discover an overcomplete dictionary of interpretable features. Our analysis reveals that the solver naturally develops features mirroring fundamental TSP concepts: boundary detectors that activate on convex-hull nodes, cluster-sensitive features responding to locally dense regions, and separator features encoding geometric partitions. These findings provide the first model-internal account of what neural TSP solvers compute before node selection, demonstrate that geometric structure emerges without explicit supervision, and suggest pathways toward transparent hybrid systems that combine neural efficiency with algorithmic interpretability. Interactive feature explorer: https://reubennarad.github.io/TSP_interp

Related papers

• Parseval Convolution Operators and Neural Networks [16.78532039510369]
  We first identify the Parseval convolution operators as the class of energy-preserving filterbanks. We then present a constructive approach for the design/specification of such filterbanks via the chaining of elementary Parseval modules. We demonstrate the usage of those tools with the design of a CNN-based algorithm for the iterative reconstruction of biomedical images.
  arXiv  Detail & Related papers  (2024-08-19T13:31:16Z)
• Convergence Analysis for Deep Sparse Coding via Convolutional Neural Networks [7.956678963695681]
  We explore intersections between sparse coding and deep learning to enhance our understanding of feature extraction capabilities.<n>We derive convergence rates for convolutional neural networks (CNNs) in their ability to extract sparse features.<n>Inspired by the strong connection between sparse coding and CNNs, we explore training strategies to encourage neural networks to learn more sparse features.
  arXiv  Detail & Related papers  (2024-08-10T12:43:55Z)
• TCCT-Net: Two-Stream Network Architecture for Fast and Efficient Engagement Estimation via Behavioral Feature Signals [58.865901821451295]
  We present a novel two-stream feature fusion "Tensor-Convolution and Convolution-Transformer Network" (TCCT-Net) architecture. To better learn the meaningful patterns in the temporal-spatial domain, we design a "CT" stream that integrates a hybrid convolutional-transformer. In parallel, to efficiently extract rich patterns from the temporal-frequency domain, we introduce a "TC" stream that uses Continuous Wavelet Transform (CWT) to represent information in a 2D tensor form.
  arXiv  Detail & Related papers  (2024-04-15T06:01:48Z)
• Vecchia Gaussian Process Ensembles on Internal Representations of Deep Neural Networks [2.186901738997927]
  For regression tasks, standard Gaussian processes (GPs) and deep neural networks (DNNs) provide natural uncertainty quantification (UQ)<n>We propose an alternative solution, the deep Vecchia ensemble (DVE), which allows deterministic UQ to work in the presence of feature collapse.<n>DVE is compatible with pretrained networks and incurs low computational overhead.
  arXiv  Detail & Related papers  (2023-05-26T16:19:26Z)
• Revisit the Algorithm Selection Problem for TSP with Spatial Information Enhanced Graph Neural Networks [4.084365114504618]
  This paper revisits the algorithm selection problem for Euclidean Traveling Salesman Problem (TSP) We propose a novel Graph Neural Network (GNN), called GINES. GINES takes the coordinates of cities and distances between cities as input. It is better than the traditional handcrafted feature-based approach on one dataset.
  arXiv  Detail & Related papers  (2023-02-08T13:14:20Z)
• Robust Training and Verification of Implicit Neural Networks: A Non-Euclidean Contractive Approach [64.23331120621118]
  This paper proposes a theoretical and computational framework for training and robustness verification of implicit neural networks. We introduce a related embedded network and show that the embedded network can be used to provide an $ell_infty$-norm box over-approximation of the reachable sets of the original network. We apply our algorithms to train implicit neural networks on the MNIST dataset and compare the robustness of our models with the models trained via existing approaches in the literature.
  arXiv  Detail & Related papers  (2022-08-08T03:13:24Z)
• Learning A 3D-CNN and Transformer Prior for Hyperspectral Image Super-Resolution [80.93870349019332]
  We propose a novel HSISR method that uses Transformer instead of CNN to learn the prior of HSIs. Specifically, we first use the gradient algorithm to solve the HSISR model, and then use an unfolding network to simulate the iterative solution processes.
  arXiv  Detail & Related papers  (2021-11-27T15:38:57Z)
• Connecting Weighted Automata, Tensor Networks and Recurrent Neural Networks through Spectral Learning [58.14930566993063]
  We present connections between three models used in different research fields: weighted finite automata(WFA) from formal languages and linguistics, recurrent neural networks used in machine learning, and tensor networks. We introduce the first provable learning algorithm for linear 2-RNN defined over sequences of continuous vectors input.
  arXiv  Detail & Related papers  (2020-10-19T15:28:00Z)
• Provably Efficient Neural Estimation of Structural Equation Model: An Adversarial Approach [144.21892195917758]
  We study estimation in a class of generalized Structural equation models (SEMs) We formulate the linear operator equation as a min-max game, where both players are parameterized by neural networks (NNs), and learn the parameters of these neural networks using a gradient descent. For the first time we provide a tractable estimation procedure for SEMs based on NNs with provable convergence and without the need for sample splitting.
  arXiv  Detail & Related papers  (2020-07-02T17:55:47Z)
• MetaSDF: Meta-learning Signed Distance Functions [85.81290552559817]
  Generalizing across shapes with neural implicit representations amounts to learning priors over the respective function space. We formalize learning of a shape space as a meta-learning problem and leverage gradient-based meta-learning algorithms to solve this task.
  arXiv  Detail & Related papers  (2020-06-17T05:14:53Z)
• Deep neural networks for inverse problems with pseudodifferential operators: an application to limited-angle tomography [0.4110409960377149]
  We propose a novel convolutional neural network (CNN) designed for learning pseudodifferential operators ($Psi$DOs) in the context of linear inverse problems. We show that, under rather general assumptions on the forward operator, the unfolded iterations of ISTA can be interpreted as the successive layers of a CNN. In particular, we prove that, in the case of LA-CT, the operations of upscaling, downscaling and convolution, can be exactly determined by combining the convolutional nature of the limited angle X-ray transform and basic properties defining a wavelet system.
  arXiv  Detail & Related papers  (2020-06-02T14:03:41Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.