PaperarXiv:2611.23085 · cs.CV · Nov 17, 2026
Transfer Once, Ask Locally: Compact Audio-Visual Memory for On-Device Question Answering
Qitong Wang, Jihui Jin, Divya Kothandaraman, Vishnu Raj, Christopher Rasmussen
Review pending. Harness Hub reviews new papers adversarially as they arrive; until then, here is the authors’ abstract.
Original abstract (authors’ text, shown)
Audio-visual question answering on lightweight user devices requires balancing speed, memory usage, privacy, and communication. One approach is to run audio-visual encoders on a server and a lightweight language decoder on the user device, keeping questions local while offloading media encoding. However, transferring and processing the full multimodal representation remains costly. We propose a training-free, query-agnostic token pruning framework that constructs a compact, reusable memory before transmission. We define video spatio-temporal blocks and aligned audio windows as players and estimate their importance through Monte Carlo Shapley valuation. We use an embedding coverage objective that combines query-independent utility with cross-modal redundancy calibration and rewards unimodal diversity through marginal coverage gains. We retain the original token embeddings of high-scoring players under a limited token budget and transfer the resulting memory once for reuse across subsequent local questions. Experiments on Video-MME, Daily-Omni, and Video-Holmes using video-SALMONN 2+ (3B) and video-SALMONN 2 Pro (4B) show that our method achieves the best or tied-best QA accuracy among the compared compression methods across all evaluated settings. On Daily-Omni with video-SALMONN 2+ (3B) at 80\% pruning, our method improves accuracy by 12.86--14.62 percentage points over these methods while achieving the lowest device-side LLM prefill and decoding latency, with up to $1.92\times$ decoding speedup at comparable embedding memory usage.
Related papers
arXiv:2611.26553 · cs.CR · Nov 19, 2026
Agents based on large language models (LLMs) can now delete files and send payments on behalf of their users. News media reports routinely describe failures of these agents (e.g., Replit's coding agent deleting a production database), but a report often only says what went wrong and rarely (in)validates the incident or finds its root cause. Incident catalogs collect these reports, and benchmarks record whether agents fail, but neither replicates an incident nor tests the cause its report might identify. In this paper, we present ROGUE-AI (Replication of Agentic Security Events: Audit and Investigation), a multi-agent framework that closes the above gap and rebuilds each reported incident offline together with a matched control that lacks only the blamed cause, and runs agents on both. Our results suggest that ROGUE-AI is able to reproduce the reported harm in 89% of its 108 attempted replications. Agents built on five current GPT, Claude, Gemini and DeepSeek models made the reported mistake in 18% to 33% of runs, against at most 1% without the blamed cause. Of the 72 incidents with real-world harm, 62 had no underlying attacker, and in these, agents did what their task asked for but missed an ordinary hazard, such as a database shared with the live site. Replications without an attacker caused more harm than replications of prompt injection for all five models. Defenses against prompt injection look for injected instructions, but these replications contain none. Further, a warning after each tool output that names the hazard and its consequence lowered harm for all five models, whereas a note of the same length that does not mention the hazard left most of the harm in place. Built from real incidents, the ROGUE-AI replications form a reusable audit framework that vendors can rerun before each new release.
Mengxiao Wang, Mahbub Alam +2
arXiv:2611.26450 · cs.AI · Nov 19, 2026
Pretrained language models (LMs) remain unreliable on exact multi-digit arithmetic without gradient fine-tuning or external tool interpreters. We introduce closed-form, zero-gradient circuit-grafting primitives that compile deterministic computation directly into pretrained LM weights using zero training examples and zero runtime hooks, while bounding per-layer normalization drift and preserving natural-language (NL) competence. Our construction combines three analytical mechanisms: (1) two-layer Mellin-GLU quadrature over bias-free SwiGLU/GeGLU layers, turning unknown per-token RMSNorm scaling into translation along a log-spaced gate grid ($>250\times$ dynamic range) with IEEE 754 error bounds; (2) cross-token scale cancellation via signed Chebyshev Indefinite-RoPE self-sink attention and memoryless SiLU-antisymmetric GatedDeltaNet shifts; and (3) axis-aligned width extension ($d \to d+K$) with $\sqrt{d/(d+K)}$ norm compensation, Base-1000 canonical BPE emission, and differential $(+254,-256)$ tied unembeddings. An automated 5-stage compiler synthesizes 4-8 digit signed $\{+,-,\times\}$ circuits in constant $\mathsf{TC}^0$ depth alongside 3-literal Boolean (100%, $N=1{,}728$; plus 42-layer Gemma 4 E4B audits) and 5-bit Majority/Parity (100%) circuits. On 6-digit arithmetic ($N=1{,}200$, BF16), IBM Granite 4.2 3B (1.07M-1.33M active non-zeros, $\le 0.036\%$) and Alibaba Qwen 3.5 4B (2.25M-2.39M non-zeros, $\le 0.057\%$) reach 99.17% and 100.00% ALU-prefix accuracy (100% with an FP32 sidecar) and 98%-100% mid-sentence resumption, outperforming 12.8K-sample LoRA (42.2% / 61.7%, 0% inline continuation). A factorized block-sparse sidecar holds A100 memory at 7.89 / 10.65 GiB (4.60 / 8.14 GiB under INT8/INT4 + FP32 sidecar) while passing all 14/14 pre-registered $>2$K-token Tango splits ($N=4{,}392$; 0/11,134 OOD triggers).
Tanmay Sah
arXiv:2611.26435 · cs.DC · Nov 19, 2026
LLM agents run long, multi-turn programs whose contexts grow at every step, and serving them is limited by KV cache capacity. To alleviate this limitation, systems use context compression techniques, which typically wait until the KV cache crosses a large threshold and then summarize large spans of history at once, either in the background while the agent progresses or on the critical path. These summarizations can be done using the same model running on the same engine, or using the same or a different model running on extra hardware. When running on the same engine, summarization competes with the requests being served for both compute and memory capacity. To maximize the cost-effectiveness of summarization, this paper presents a new serving system called {\em Aggron}. Aggron summarizes agentic requests at every turn and selectively summarizes only certain parts of the KV cache. As a result, Aggron retains accuracy and keeps summarization requests lightweight. Moreover, Aggron launches summarization requests from a separate {\em Summarization Queue} while the corresponding agentic request whose KV cache we want to compress is stalled in the main request queue due to head-of-line stalls. The Summarization queue utilizes compute and memory capacity left over by the head-of-line stalls in the main queue. Further, to maximize efficiency, the server issues summarizations only when it predicts that they can complete before the corresponding agentic request will finally execute. In our evaluation, we compare Aggron to state-of-the-art asynchronous summarization. At high load, Aggron reduces P99 program latency by $27$--$31\%$, and raises throughput by $1.34$--$1.40\times$.
Rishabh Ravi, Tony Hong +6