$PRUN pruning

Latest papers

1. Your Transformer Is Secretly Elastic cs.LG · 14 Oct · 3 hits

   …MLP neuron and attention head jointly, on a common scale, by minimizing the model's loss under soft pruning across many sparsity targets, all at once. Across 27 vision and text transformer encoders ranging…

2. Selective Growing Q-Networks (SGQN): An Analysis of Adaptive Action Grids in Decoupled Q-Learning cs.LG · 14 Oct · 2 hits

   …find that it does not. Growth activates every candidate bin and behaves like uniform growth, and removing pruning changes the return by at most 1.1%. Two other choices matter: starting from a coarse grid, since…

3. Where Should the Gaussians Go? Reusable Label-Aware Allocation for Budget-Constrained Gaussian Splatting of Cinematic CT cs.CV · 14 Oct · 1 hit

   …patients at 10K-100K Gaussians, both raise vessel-masked PSNR over uniform allocation by 4.5-7.1dB and over pruning a dense voxel-aligned model by 1.6-3.3dB. The neural prior does not differ significantly from…

4. STP-GUI: Spatially-Aware Token Pruning for Efficient Web-GUI Agents cs.CV · 14 Oct · 4 hits

   …histories introduce excessive visual tokens, causing substantial inference overhead. Existing token pruning methods, mainly designed for natural images, often select tokens independently based on importance…

5. TreeSpark: Learning to Draft the Trees You Verify for Speculative Decoding cs.CL · 14 Oct · 1 hit

   …It constructs parent-conditioned candidate trees from one backbone pass and replays their expansion, pruning, and budgeted packing during training. Survival negative log-likelihood weights retained edges by…

6. KANLab: TensorFlow Playground for Kolmogorov-Arnold Networks cs.LG · 14 Oct · 1 hit

   …manual disruptions. Furthermore, user can explore phenomena such as activation variance, interactive pruning, and re\-/training within a compact 2D machine learning setting. KANLab is open-source and available…

7. Towards Dynamic Top-$k$ Simple Shortest Paths cs.DS · 14 Oct · 1 hit

   …that generates candidate paths and combinable with known preprocessing techniques, as well as with an online pruning rule that we formalize. While a worst-case update is asymptotically comparable to recomputing from…

8. Choosing the Number of Attention Heads from Intrinsic Dimension: A Small-Scale Study cs.CL · 14 Oct · 1 hit

   …for the 10-head model. Training 10 heads from the start also slightly outperforms training 32 heads and pruning to 10 (0.6 points, $p=0.008$). The evidence is narrow: two classification tasks, one small…