HiVe-Mem cuts multimodal agent memory tokens by about 90%
A graph of textual and visual anchors plus a trainable evidence selector improved three memory benchmarks across base models.
Stories per day · 30 days
7 RSS
Relatedrag9embeddings6agent-memory4benchmarks4llm-agents4memory4qwen4benchmark3language-models3nemotron3nvidia3reinforcement-learning3reranking3decision-models2enterprise-agents2locomo2long-context2multi-agent2
104 stories, newest first · one card per story
Stories tagged #retrieval
Saidzoda fine-tune lifts Belebele recall for four Central Asian languages
Belebele recall@1 rises 14.4 to 26.0 points, to 75.2%–77.7%, while English and Russian recall fall.
Open-Weight Decision Models Screen Web Passages Against Injection Attacks
A Kev-4B gate reduces attack success from 27.9% to 5.7% on a 1,000-question benchmark.
Box to embed Nvidia Nemotron in content retrieval
Enterprise Advance customers get it early next year; Nvidia expects faster queries and 90% fewer tokens.
Trainable [DUMP] prefix tokens cut interference in retrieval embeddings
The method leaves architecture, pooling, and contrastive loss unchanged and reports gains on retrieval and long-context benchmarks.
ArcRAG enables construction-agnostic graph retrieval
Query-local graph interface induction allows fixed retriever to work on differently built knowledge graphs.
Attention competition predicts context rot in retrieval
Filler tokens maintain accuracy up to 32K tokens in retrieval but rival statements multiply loss by 9-110 across five open models.
TRIO aligns three agents to refine retrieval-augmented generation
Query, judge, and answer agents are trained with separate rewards, including voting labels and curriculum matching.
DecisionRank on-policy distillation beats generative teacher on retrieval tasks
Item-space OPD lets 1.7B DecisionRank surpass 27B LLM teacher on TREC DL while running 24.5x faster.
DecisionRank trains decision model rerankers
DecisionRank achieves SOTA on TREC DL and BEIR with a 1.7B model that surpasses its 27B generative teacher.
Paper treats long-context retrieval as value-indexed local traversal
On MiniGrid Memory, a five-token traversal head reached 0.95 mean return after 0.59M steps, ahead of absolute addressing.
Paper ties RL memory retrieval and credit to value of information
A learned world model estimates that quantity from archived rollouts, defining an optimal retrieval set and a TD-target floor.
MemSteer builds memory-optimization labels from unlabeled user histories
MemSteer-Opt lifts ATM-Bench Question Score by 13.4 points over Base-BM25 without changing model weights.
Glance reranks passages from the generator's shared prefill
A resident 9B backbone adds about 0.1 GiB per skill and beats bge-reranker-v2-m3 on SciFact and NFCorpus after 39 minutes of fine-tuning.
llms.txt audit finds no isolated effect on AI search security
A 120-response pilot found attacker-target adoption in 21 of 24 raw adversarial replies when evidence was supplied.
Skill-Uni jointly trains LLM skill retrieval and use
Authors report average gains of 66.59% on Qwen3-4B and 34.82% on LLaMA3.1-8B across five benchmarks.
SKIM stores dialogue turns inside frozen language-model parameters
The model writes self-generated questions as retrieval keys and later answers user questions from parameters alone, with no external store.
TrawMem organizes agent memory by task role
On LoCoMo and MemGallery, authors report macro-F1 gains of 2.23–3.14 points and fewer context tokens than compared memory systems.
DecisionRAG retrieves passages with token-free decision models
Without task-specific training, it beats dense retrieval plus an 8B reranker by 13.8 R@1 on MultiDoc2Dial.
Visual memory raises recall but lowers precision in long video
On VMem-Bench, retrieval lifted entity recall from 0.559 to 0.628-0.693 while precision and forbidden-entity avoidance fell.
HerHealthGPT grounds reproductive-health answers in retrieval and clarification
On 540 messages from 90 cases, retrieval lifted concern-category accuracy from 54.1% to 59.3%.
Paper says RAG poison filters rely on construction shortcuts, not content
Stripping PoisonedRAG's question prefix drops reimplemented RAGuard from AUROC 0.953 to 0.758 and raises its miss rate to 0.763.
Retrieval heads stay sparse and shared across MLLM modalities
Ranking training on those heads raises Qwen3-VL-8B page Recall@1 on MMDocIR by 6.03 points.
Cohere releases Rerank 5 with passage-level context focus
The captured page is only a cookie notice, so no release details beyond the title are available.