Web This site Trusted
30 result(s) for “okapi bm25 site hotchips pp site can kernel org site who unesco org site have apache site:arxiv.org” of 54 matched · 548 ms
Scope: Nishi’s own index (our crawl + Common Crawl) — not the whole live web. Some sites block all automated crawlers (anti-bot walls), so their pages may be absent here even though we never filter lawful content.
Freshness: newest result fetched 26 d ago · median age of these results 26 d · 1 of 30 carry a fetch time
The query asks the model to recall the topic of an earlier conversation about Tesla; the gold answer is anchored in one specific session about Pooja’s dig site
from the top result below
https://arxiv.org/html/2608.02613v1
We evaluate five open-weight readers with Vanilla context, BM25-RAG, Oracle retrieval, Memobase, and MemSearch as memory backends
https://arxiv.org/html/2501.13956v1
Current approaches using RAG have focused on broad domain knowledge and largely static corpora—that is, document contents added to a corpus seldom change
https://arxiv.org/html/2608.03048v1
for large language models, as recent recurrent-memory approaches face two inherent challenges: sequential chunk-wise updates can overwrite early critical evidence with later irrelevant content, and serial inter-chunk dependencies limit parallelism and cause latency to increase with context length
https://arxiv.org/html/2609.03027v1
EPIC can achieve 𝒪(NlogN)\mathcal{O}(N\log N) computational scaling for dense aperture arrays and has been demonstrated on the Long Wavelength Array (LWA) station in Sevilleta (USA) using Graphics Processing Unit (GPU)
https://arxiv.org/html/2609.02852v1
However, various strands of evidence indicate that an LLM’s externalized linguistic outputs and mechanistically-extracted linguistic features can be an unreliable lens for understanding internal model computation
https://arxiv.org/html/2607.27099v1
org perpetual non-exclusive license
arXiv:2607
https://arxiv.org/list/cs.AI/new
We demonstrate that it can self-assemble functional circuits from scratch and rapidly re-route logic around permanent, previously unseen hardware faults
https://arxiv.org/abs/2608.03883
Our systems combine sparse BM25 retrieval, evidence-aware rescoring for self-referential symptom reports, embedding-based reranking, query-prototype expansion, and LLM-based reranking
https://arxiv.org/abs/2608.03860
The pipeline combines sentence-window chunking, BM25, BGE-M3 dense retrieval, reciprocal rank fusion, optional cross-encoder reranking, and grounded answer generation
https://arxiv.org/abs/2104.08663
Benchmark for Zero-shot Evaluation of Information Retrieval Models, by Nandan Thakur and 4 other authors
View PDF HTML (experimental)
Abstract:Existing neural information retrieval (IR) models have often been studied in homogeneous and narrow settings, which has considerably limited insights into their out-of
https://arxiv.org/list/cs.IR/recent
02048 [pdf, html, other]
Title: SmartGR: Hierarchy and Beam-Aware Knowledge Distillation for Generative Recommendation
Ziheng Zhang, Yu Cui, Bohao Wang, Yong He, Chao Yu, Chuan Yuan, Wujie Sun, Can Wa
https://arxiv.org/list/cs.IR/new
Our results show that skill representation itself matters, and that simply preserving the structure already present in skill files can substantially improve retrieval
https://arxiv.org/list/math.ST/2024-03
00186 [pdf, html, other]
Title: Warped Kernel Estimator for I
https://arxiv.org/list/cs.CL/new
LG)
Half a million cuneiform clay tablets survive in museums worldwide, yet modern users can neither read nor write in the world's oldest writing system, leaving a 4,000-year cultural barrier that existing NLP tools have only partially addressed
https://arxiv.org/html/2407.16741v3
In particular, AI agents, systems that can perceive and act upon the external environment, have recently received ever-increasing research focus
https://arxiv.org/abs/2608.12700
A kernel can pass that test and still be silently wrong
https://arxiv.org/abs/2204.04218
In many situations, computed tomography (CT) or magnetic resonance imaging (MRI) techniques capture several scans (modes) during a single investigation, which can jointly be used (in a multimodal fashion) to further boost the quality of super-resolution results
https://arxiv.org/html/2608.02851v1
org perpetual non-exclusive license
arXiv:2608
https://arxiv.org/list/cs.DC/new
SE)
Automating GPU kernel optimization remains difficult in practice: generated variants can violate correctness constraints, runtime measurements are noisy, and search often stalls early
https://arxiv.org/html/2607.22511v2
A statement can be free of sorry and still be mathematically uninformative: a definition may reduce to True; or an unproven step may be introduced as an axiom that the kernel accepts
https://arxiv.org/list/cs.DB/new
We propose an architectural pattern for such systems, a trusted kernel with a generative shell, resting on one invariant: a component that can fabricate may influence which question the system answers, never which value it returns
https://arxiv.org/html/2508.20101v1
The model combines classical GARCH(pp, qq) dynamics with spatially correlated innovations and spatially varying parameters, estimated using local likelihood methods
https://arxiv.org/abs/2608.02851v1
The design relies on two affine Volterra transformations involving both Fredholm- and Volterra-type integral terms, which results in a four-PIDE kernel equations
https://arxiv.org/list/cs.OS/new
Applications may therefore slow down before the kernel kills a process
https://arxiv.org/abs/2512.09590
Using an extension of the exponential-affine transformation formula for those processes, we establish in the long run the existence of limiting distributions, which (unlike in the case of classical affine diffusion processes) may depend on the initial state of the process, unless the Volterra kernel coincides with the
https://arxiv.org/abs/2608.01651
We present Bole, a kernel--runtime co-design that enables efficient tree speculation for hybrid-attention LLMs
https://arxiv.org/abs/2609.14858
Our key insight is that accumulated discovery history can serve as a replay simulator over the realized search space
https://arxiv.org/abs/1202.3424
, dust particles), but may also be 1-d or 2-d periodic arrays of "target unit cells", which can be used to study absorption, scattering, and electric fields around arrays of nanostructures
https://arxiv.org/abs/2504.02246
C* extends C with verification capabilities, powered by a symbolic execution engine and an LCF-style proof kernel
https://arxiv.org/list/cs.AR/new
Our analysis shows that, despite its limited write performance, HBF can significantly improve the batch size, throughput, and flexibility of LLM-serving systems while reducing the minimum GPU requirements, but realizing these benefits critically depends on sustaining HBM-comparable read bandwidth and requires