The paper investigates how the choice of retrieval encoder affects cache‑based test‑time adaptation for CLIP. By keeping the memory fixed and varying the retrieval space across sixteen encoders, the authors show that retrieval space can dramatically alter performance, with gains ranging from +0.44 to +19.7 points on ImageNet‑A. They introduce MARC, a training‑free system that pairs frozen CLIP with DINOv2‑B for retrieval, achieving superior out‑of‑distribution accuracy and efficiency compared to prior methods.
By Mahir Shahriar Tamim, Md. Samiul Alim, Azmine Toushik Wasi, Shahriyar Zaman Ridoy, Meharun Nesa, Mohammad Abu Yousuf, Alex Lamb, Mohammad Ali Moni
The paper investigates how much semantic information is lost when frozen foundation models are combined for few‑shot 3D segmentation. By varying the number of retained semantic alternatives before fusion, the authors show that keeping the full distribution of class scores yields higher harmonic‑mean IoU than collapsing to a single class. Experiments on ScanNet200 and ScanNet++ confirm that full‑distribution fusion consistently outperforms top‑1 and other operators, and that most useful information is recovered by retaining a compact set of plausible alternatives.
By Silas Kwabla Gah, Ebenezer Owusu
arXiv:2606. 25432v1 Announce Type: new Abstract: Inference efficiency is typically pursued by shrinking the model: distillation, pruning, quantization, and sparse routing each lower per-token cost while treating token count as fixed.
By DatologyAI, :, Matthew L. Leavitt, Siddharth Joshi, Haoli Yin, Rishabh Adiga, Haakon Mongstad, Alvin Deng, David Schwab, Bogdan Gaza, Ari Morcos
arXiv:2303. 15619v2 Announce Type: replace-cross Abstract: The choice of \emph{which} tokens to mask is a central, under-examined design decision in masked language modeling (MLM).
By Muhammed Shahir Abdurrahman, Hashem Elezabi, Bruce Changlong Xu
arXiv:2411. 16073v4 Announce Type: replace-cross Abstract: Inspired by the Well-initialized Lottery Ticket Hypothesis (WLTH), we introduce Soft-TransFormers (Soft-TF), a continual learning framework that adapts a frozen pre-trained Transformer through task-specific soft subnetworks: real-valued multiplicative masks over the query, key, value, and output projections of selected self-attention layers.
By Haeyong Kang, Chang D. Yoo
Inference efficiency is typically pursued by shrinking the model: distillation, pruning, quantization, and sparse routing each lower per-token cost while treating token count as fixed. But output length has been inflating, and it is precisely the component the standard toolkit leaves untouched.
The study investigates how long‑video language models decide which frames to keep, compress, and reuse, testing each decision in isolation across six selection rules, three benchmarks, and two answering models. It finds that selecting frames based on queries yields the biggest performance boost, that halving spatial resolution costs little, and that reallocating saved tokens to more compressed frames can further improve accuracy. The work also highlights the importance of a unified evaluation harness to avoid misleading comparisons.
By Prakhar Khatri
The paper evaluates coreset selection methods by incorporating both selection and training time into a unified wall‑clock budget, using a standardized benchmark across four datasets and multiple selectors. Across numerous budget anchors, simple random or full‑data training consistently outperforms sophisticated selectors, and selection costs are dominated by a full‑dataset scan that cannot be amortized. The study also identifies when subset reuse can justify selection and reports several correctness fixes in a popular codebase.
By Yangze Liu, Zhongyi Han
The paper investigates KV‑cache eviction strategies for sparse‑attention models, showing that selecting the largest attention weights is nearly optimal—closing only a median 2–5 % of the gap to full attention. It further demonstrates that differences in performance between eviction methods largely stem from memory usage, with the new training‑free ContourKV allocator outperforming state‑of‑the‑art methods in most pairwise comparisons while matching their byte‑efficiency.
By Jack Shi, Jerry Gu
arXiv:2606. 01502v1 Announce Type: cross Abstract: Frontier LLMs increasingly decide what a query attends to with a sparse-attention indexer that picks a few KV-cache blocks per query: attention's unit is now a small, reusable chunk.
By Bole Ma, Jan Eitzinger, Harald K\"ostler, Gerhard Wellein
arXiv:2609.03949v2 Announce Type: replace-cross
Abstract: A long-lived KV cache must be compressed before the queries that will read it exist. Selection by observed attention collapses there: on a No...
By WenJie Fan
arXiv:2608.21098v1 Announce Type: new
Abstract: Fusing prior knowledge with data-driven learning is attractive where data is scarce, yet no controlled account says when it helps, is redundant, or har...
By Ahmad AlMughrabi, Albert Clop, Benjamin Busam, Ricardo Marques, Petia Radeva