LoRA, PEFT, instruction tuning and domain adaptation — adapting a pretrained model without paying to train one.
arXiv:2609.16059v1 Announce Type: cross Abstract: Multimodal instruction following (MMIF) is crucial for building generalist agents. However, current training paradigms rely heavily on Supervised Fin...
VOR-Bench is a new benchmark for video object removal that addresses shortcomings in current evaluation methods by providing a dataset with paired edited videos and graffiti masks, a realistic motion-capable paired-video acquisition framework (rMPAF), and a perception-driven scoring model (VOR-MDSM). The dataset includes diverse data from model-generated, tool-rendered, and camera-captured sources, ensuring robust real-world assessment. Experiments show that VOR-Bench’s evaluation results correlate strongly (ρ > 0.9) with human subjective judgments, bridging the gap between traditional metrics and human preference.
EgoPathBench is a new dataset and benchmark that tests zero‑shot egocentric waypoint decision‑making in vision‑language models. Each task presents an egocentric RGB image, a natural‑language goal, and numbered visible waypoints, and models must return traversable candidates or an ordered route. The benchmark evaluates candidate feasibility, edge legality, and goal arrival under point‑agent or embodied geometry, covering 31,852 training, 1,345 validation, and 1,111 benchmark questions. "whyItMatters":"The benchmark reveals that current VLMs perform poorly on integrated navigation tasks, highlighting a gap in spatial intelligence that can be addressed by fine‑tuning with the released training data."
The paper introduces a framework that uses Large Language Models (LLMs) to generate synthetic time‑series data for manufacturing processes. By fine‑tuning pre‑trained LLMs on manufacturing instructions and applying Retrieval Augmented Generation (RAG), the method enhances data diversity and realism. Evaluation against traditional models such as ARIMA and LSTMs shows that the LLM‑driven approach produces higher‑quality synthetic data, better capturing temporal dependencies and improving downstream anomaly detection performance.
arXiv:2607.09709v2 Announce Type: replace Abstract: Post-training a code generator against a learned judge can optimize proxy features that raise the score without improving the artifact. We study th...
The paper introduces DiaWhisper-DPO, an end‑to‑end model that fine‑tunes Whisper-large-v3 with LoRA and a frame‑level role head to transcribe and attribute utterances in clinical interviews. It further refines the system using failure‑mined preference optimization (DPO) that leverages genuine decoding failures as rejected completions, eliminating the need for human preference data. On the DAIC‑WOZ dataset, DiaWhisper‑DPO attains 0.973 role accuracy and 0.119 DER, outperforming cascaded baselines by 72% and dramatically reducing seed variation, while also improving performance on the cross‑lingual PDCH‑HAMD dataset.
arXiv:2609.17019v1 Announce Type: new Abstract: While Chain-of-Thought (CoT) reasoning has been proven to be effective, it often leads to overthinking, resulting in computational overhead, inference...
The paper introduces CoER, a framework that defends language‑model agents against adaptive indirect prompt injection (IPI) by employing attacker‑defender co‑evolution and refinement. CoER models IPI as a general‑sum Markov game, uses Co‑PPO to maintain historical opponent populations, and fine‑tunes defenders only on verified safe demonstrations. In experiments across seven domains, CoER cuts attack success from 38.5% to 0.2% while boosting task utility from 63.2% to 76.3%.
GrowMTP is a method that trains a draft head entirely within the reinforcement learning (RL) loop, using supervision from the RL verification step and a rollout distribution that is narrower than pretraining. By detaching draft‑head updates from the policy backbone, it enables online training of the draft head from scratch. Experiments on Qwen3‑4B, MiMo‑7B‑SFT, and Qwen3.5‑4B‑Base show rollout speedups ranging from 1.36× to 2.13× and overall end‑to‑end speedups from 1.20× to 1.60×, making it a modular acceleration component for RL frameworks lacking pretrained draft heads.
The paper investigates whether natural language inference (NLI) can be performed using only interpretable, graph-based representations of evidence. It introduces a pipeline that decomposes sentences into atomic propositions, maps them to ConceptNet triples, and feeds three graphs—premise, hypothesis, and a retrieved ConceptNet subgraph—into a fine‑tuned 0.8‑billion‑parameter language model. On the SNLI dataset the graph‑only model reaches 89.7% accuracy, close to a text‑based baseline, while on ANLI it matches RoBERTa‑large on rounds R2 and R3 but lags on R1, illustrating a trade‑off between interpretability and performance.
arXiv:2609.16032v1 Announce Type: cross Abstract: Diffusion-based artifact detectors score whole-slide image patches by reconstruction error under a model fine-tuned on clean tissue. We show that con...
arXiv:2609.17026v1 Announce Type: new Abstract: Continual learning must balance the learning of new knowledge with the retention of previously learned knowledge to incrementally learn tasks from a da...
The paper evaluates audio‑visual predictive turn‑taking models trained on clean data when applied to a noisy cocktail‑party scenario derived from the AVCocktail dataset. Results show a consistent performance drop—up to 38% relative in weighted F1—across both audio and visual modalities, with fine‑tuning improving robustness but varying by modality and pre‑training data size. The study highlights differing generalisation and adaptation abilities of audio versus visual inputs and underscores the need for robust modelling strategies in noisy human interactions.
FLAT (Flexible‑Length Aligned Transmodal representations) is a joint multimodal pre‑training framework that learns a shared encoder for images and text, producing 1‑D continuous embeddings that can be directly used by downstream generative decoders. By combining contrastive alignment with bidirectional cross‑modal generative objectives, FLAT yields representations that are both discriminative and generative, enabling cross‑modal retrieval and generation with a single pre‑training stage. The model achieves strong performance on T2I generation (GenEval 71.1), image captioning (BLEU‑4 40.5, CIDEr 138.6), and retrieval tasks (Recall@5 86.8/75.8 on MS‑COCO, 98.3/93.6 on Flickr30K), and supports linear interpolation, latent space arithmetic, and zero‑shot composed retrieval.
HoloAegis is a minimally parametric topological inference framework that uses frozen representations to map text onto the unit sphere and makes decisions via Gibbs‑Boltzmann free‑energy differences over pre‑computed anchor centroids. On a frozen three‑benchmark protocol, it matches WildGuard‑7B on toxicity, outperforms it on harmful behaviors, but underperforms on oversafety detection, while ShieldGemma‑2B fails on indirect harms. The study demonstrates that geometric guardrails can substitute for LLM judges in some cases and must defer to them in others, with anchor banks reducing score variance and boundary displacement.
arXiv:2609.16859v1 Announce Type: cross Abstract: To train handwritten text recognition systems we need word images and their corresponding transcriptions, and these transcriptions are produced manua...
Multi-encoder fusion extends Large Audio-Language Models (LALMs) beyond speech-centric recognition, but selecting encoders via intuition or exhaustive search often introduces redundant representations...
Speech Large Language Models (Speech-LLMs), typically built from a pre-trained speech encoder, a modality projector, and an LLM fine-tuned with Low-Rank Adapters (LoRA), have shown strong Automatic Sp...
Table understanding is a core task in document intelligence, encompassing two key subtasks: table reconstruction and table visual question answering (TabVQA). While recent approaches predominantly rel...
While draft-model-free speculative decoding offers a promising path to efficient LLM inference, it is frequently constrained by stale draft candidates and the high computational cost of the verificati...