Consistency-based defenses for retrieval-augmented generation (RAG) rest on an assumption that a sparse adversary breaks: a poisoning campaign may modify a negligible fraction of the corpus, yet retrieval concentrates its passages into the reader's context. Across four Natural Questions retrieval seeds, at least one in...
Attention-Scoped Guidance (ASG), a sampler wrapper that reads a soft support map from the instruction attention that the editor already computes, then weakens text guidance where support is low and strengthens image anchoring where support is high.
Zeyan Li, Wei Zhou, Hadi Amirpour et al.· 1 citation
CPUNeSy, a serving architecture that controls model writes to symbolic state via a task-defined predicate interface and certificate gate, abstaining when grounding passes disagree is introduced, showing deterministic execution drives most accuracy recovery on derivation-heavy tasks.
Zeyan Li, Si-Yuan Qiu, Shuai Zhao et al.· 0 citations
Retrieval-augmented language models are expected to answer from the retrieved evidence, but in practice they often keep answering when that evidence is missing. We trace this behavior to the training signal: answer-focused fine-tuning assigns no target to unsupported contexts, so it cannot distinguish a reader that abs...
Zeyan Li, Qi-Rong Guo, Si-Yuan Qiu et al.· 0 citations
SemRD-V2X is an operational neural proxy that combines exact-budget BEV support selection, pointwise channel compression, and masked shared-weight reconstruction before standard fusion before standard fusion, and position closure fidelity as both an analytical lens and an actionable design principle for communication-e...
Low-rank adapters (LoRA) make it cheap to fine-tune a large language model once per task, but combining several independently trained adapters into one model remains difficult: merging the updates in weight space causes interference, retraining on all task data is expensive, and routing between separate adapters gives...
Zeyan Li, Pan-Qi Yang, Qi-Rong Guo et al.· 0 citations
Backbone-Adaptive Evidence Routing (BAER), which adapts the evidence mechanism while preserving candidate symmetry, achieves the highest test accuracy among the compared methods in all eight conditions, with full prediction coverage and gains over the strongest external baseline.
BridgeMem is introduced, which estimates the value of explicit dyadic transition modeling as a residual added to the log scores of a frozen full-vocabulary forecaster, and shows the value of explicit dyadic transition modeling in temporal knowledge graph forecasting.
Chain-of-thought (CoT) instructions can distort multiple-choice VLM evaluation when a scorer appends a reasoning cue but reads answer-label logits before the model generates any rationale. We call this CoT-prefix scoring. On ScienceQA, Qwen2.5-VL-7B drops from 80.76% to 45.48%, and across five option-content permutatio...
Knowledge editing changes what a model knows by modifying parameters so that a requested fact updates while unrelated behavior is preserved. This is usually treated as a write problem, but editing also involves an address problem: deciding which hidden states should receive the new residual. An update that activates to...
Zeyan Li, Hu Xu, Jian-Feng Xu· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.