Tag Archives: search

NEW PAPER: Winning the Lottery by Preserving Network Training Dynamics with Concrete Ticket Search

T. Arora and C. Teuscher, Winning the Lottery by Preserving Network Training Dynamics with Concrete Ticket Search, 2025 Under review. https://arxiv.org/abs/2512.07142

Abstract:

The Lottery Ticket Hypothesis asserts the existence of highly sparse, trainable subnetworks (‘winning tickets’) within dense, randomly initialized neural networks. However, state-of- the-art methods of drawing these tickets, like Lottery Ticket Rewinding (LTR), are computationally prohibitive, while more efficient saliency-based Pruning-at-Initialization (PaI) techniques suffer from a significant accuracy-sparsity trade-off and fail basic sanity checks. In this work, we argue that PaI’s reliance on first-order saliency metrics, which ignore inter-weight dependencies, contributes substantially to this performance gap, especially in the sparse regime. To address this, we introduce Concrete Ticket Search (CTS), an algorithm that frames subnetwork discovery as a holistic combinatorial optimization problem. By leveraging a Concrete relaxation of the discrete search space and a novel gradient balancing scheme (GRADBALANCE) to control sparsity, CTS efficiently identifies high-performing subnetworks near initialization without requiring sensitive hyperparameter tuning. Motivated by recent works on lottery ticket training dynamics, we further propose a knowledge distillation-inspired family of pruning objectives, finding that minimizing the reverse Kullback-Leibler divergence between sparse and dense network outputs (CTSKL) is particularly effective. Experiments on varying image classification tasks show that CTS produces subnetworks that robustly pass sanity checks and achieve accuracy comparable to or exceeding LTR, while requiring only a small fraction of the computation. For example, on ResNet-20 on CIFAR10, CTSKL produces subnetworks of 99.3% sparsity with a top-1 accuracy of 74.0% in just 7.9 minutes, while LTR produces subnetworks of the same sparsity with an accuracy of 68.3% in 95.2 minutes. However, while CTS outperforms saliency-based methods in the sparsity-accuracy tradeoff across all sparsities, such advantages over LTR emerge most clearly only in the highly sparse regime.

New Paper: Proximal Policy Optimization for Radiation Source Search

Proctor, P.; Teuscher, C.; Hecht, A.; Osiński, M. Proximal Policy Optimization for Radiation Source Search. Journal of Nuclear Engineering, 2:368-397, 2021. https://doi.org/10.3390/jne2040029

Rapid search and localization for nuclear sources can be an important aspect in preventing human harm from illicit material in dirty bombs or from contamination. In the case of a single mobile radiation detector, there are numerous challenges to overcome such as weak source intensity, multiple sources, background radiation, and the presence of obstructions, i.e., a non-convex environment. In this work, we investigate the sequential decision making capability of deep reinforcement learning in the nuclear source search context. A novel neural network architecture (RAD-A2C) based on the advantage actor critic (A2C) framework and a particle filter gated recurrent unit for localization is proposed. Performance is studied in a randomized 20×20 m convex and non-convex simulation environment across a range of signal-to-noise ratio (SNR)s for a single detector and single source. RAD-A2C performance is compared to both an information-driven controller that uses a bootstrap particle filter and to a gradient search (GS) algorithm. We find that the RAD-A2C has comparable performance to the information-driven controller across SNR in a convex environment. The RAD-A2C far outperforms the GS algorithm in the non-convex environment with greater than 95% median completion rate for up to seven obstructions.