The paper introduces a two‑stage framework for recognizing depression symptoms at the sentence level. First, a contrastively fine‑tuned sentence encoder generates a symptom candidate for each sentence. Then, a fine‑tuned language model verifies the candidate’s presence or absence by comparing the sentence, its context, and a diagnostic definition, ensuring the model’s judgment aligns with that definition before responding.
By Weiming Li, Catarina Barata, Miguel Constante, Joao Sanches
The study investigates whether language models trained specifically on Cantonese better predict human reading patterns by comparing eye-tracking data with information-theoretic metrics derived from several models. Two adaptation contrasts were examined: a lightly adapted CKIP GPT-2 Tiny versus its Cantonese derivative JED351, and a heavily adapted Qwen2.5-7B versus CantoneseLLM-7B. Results show that the extensively Cantonese-trained CantoneseLLM-7B consistently outperforms others on lexical surprisal and joint metrics, while entropy reduction favors the less adapted CKIP model, indicating that deeper Cantonese-specific training can improve predictive alignment but that rankings vary by metric.
By Ziqi Zhang, Emmanuele Chersoni, Mohammad Momenian
The paper investigates why post‑training boosts reasoning more than perception in vision‑language models. Using a diagnostic framework with synthetic tasks, it finds a perception‑reasoning asymmetry: supervised fine‑tuning suffers from token imbalance, while reinforcement learning suffers from reward coupling. The authors propose reweighting losses and perception‑aware rewards, achieving up to 18.2‑point and 6.0‑point gains respectively, and show that these methods also improve real‑world visual reasoning by up to 3.3 points.
By Xueqing Wu, Yu-Chi Lin, Kai-Wei Chang, Nanyun Peng
The paper presents a method for improving markerless 3D pose estimation in infants by cross‑model distillation. Using unannotated infant video, a frozen Sapiens 2 pose model provides dense pseudo‑labels that guide fine‑tuning of the SAM 3D Body model. On a held‑out dataset of eleven infants, the fine‑tuned model shows significant gains in 2D keypoint accuracy and 3D joint error compared to the original SAM 3D Body model.
By R. James Cotton, Divya Joshi, Colleen Peyton
The study evaluates fundus-specific foundation models (FM) for detecting diabetic macular edema (DME) in retinal images. It compares two popular FM—RETFound and FLAIR—against a lightweight EfficientNet-B0 backbone across multiple datasets (IDRiD, MESSIDOR-2, and OCT-and-Eye-FundusImages). Results indicate that FM do not consistently outperform fine‑tuned CNNs; EfficientNet-B0 often matches or exceeds FM performance, with FLAIR being the most competitive FM.
By Franco Javier Arellano, Jos\'e Ignacio Orlando
The paper presents a lightweight method to adapt general‑purpose vision‑language models (VLMs) for multispectral and synthetic aperture radar (SAR) image understanding. By rendering each observation as five optical views and one SAR view, naming them in the prompt, and applying LoRA to the language network and selected visual transformer blocks, the authors enable VLMs to process band composites, spectral indices, and radar backscatter without retraining a new foundation model. On a balanced six‑class land‑cover benchmark from BigEarthNet‑v2, the adapted Qwen3‑VL achieves a micro F1 of 0.8275, and the same protocol improves four other VLMs and transfers to flood verification and captioning tasks.
"whyItMatters":"The study shows that existing VLMs can be repurposed for multispectral and SAR tasks through simple input rendering and compact LoRA adaptation, avoiding the need for dedicated encoders and domain pretraining."
By Shanji Liu, Kelu Yao, Junxiao Xue, Chenghui Lv, Xiangyang Miao, Yekai Huang, Yaying Chen, Chao Li
TikZilla is a new approach to generating TikZ code from textual descriptions, built on a larger, higher‑quality dataset called DaTikZ‑V4 that includes LLM‑generated figure descriptions. The method uses a two‑stage pipeline: supervised fine‑tuning of small Qwen models (3B and 8B) followed by reinforcement learning with an image encoder that provides semantically faithful reward signals. Human evaluations show that TikZilla outperforms its base models by 1.5–2 points on a 5‑point scale, beats GPT‑4o by 0.5 points, and matches GPT‑5 in image‑based tests while remaining much smaller.
By Christian Greisinger, Steffen Eger
UniToolCall introduces a unified framework for tool-use in large language model agents, standardizing toolset construction, dataset generation, and evaluation. The framework aggregates over 22,000 tools and creates a hybrid training corpus of more than 390,000 instances by combining ten public datasets with synthetically generated, structurally controlled trajectories. It models diverse interaction patterns—single‑hop vs. multi‑hop, single‑turn vs. multi‑turn, serial vs. parallel execution—and adds an Anchor Linkage mechanism to enforce cross‑turn dependencies, while converting seven public benchmarks into a common Query–Action–Observation–Answer format for fine‑grained evaluation.
By Yijuan Liang, Xinghao Chen, Yifan Ge, Ziyi Wu, Hao Wu, Changyu Zeng, Wei Xing, Xiaoyu Shen
The paper introduces a method that adapts Contextual Decomposition for Transformers (CD‑T) to discover neural circuits without relying on clean counterfactuals, using label‑balanced activation means and task‑directional relevance scoring. These circuits are then employed in Circuit‑Targeted Supervised Fine‑Tuning (CT‑SFT), which restricts parameter updates to task‑relevant heads and LayerNorm, leading to competitive performance on low‑resource language adaptation tasks such as NusaX cross‑lingual sentiment transfer. CT‑SFT consistently avoids catastrophic forgetting and preserves source‑language and related‑task performance, offering a more controlled alternative to global fine‑tuning, as further validated on the XNLI benchmark.
By Khumaisa Nur'aini, Ayu Purwarianti, Alham Fikri Aji, Derry Wijaya
GlyphAnchor is a new method that improves visual text rendering in image generation and editing models by adding lightweight glyph patch conditions anchored to the target image’s positional encoding. The approach is trained with staged supervised finetuning and text-aware post‑training, and it works with both text‑to‑image and image‑editing diffusion transformers. Experiments on various backbones and the newly introduced InfoTextBench benchmark show that GlyphAnchor consistently enhances text fidelity while maintaining overall image quality, especially for long, complex, or densely arranged text and rare characters.
By Qiang Xiang, Shuang Sun, Binglei Li, Yibo Chen, Xu Tang, Yao Hu, Junping Zhang
TaRA: Training-Aware Low-Rank Adaptation Initialization proposes a new way to initialize LoRA by aligning the gradients of low‑rank factors with those of the full‑rank weight matrix. This approach directly incorporates training dynamics, improving gradient fidelity at the start of fine‑tuning while adding negligible computational cost. Experiments on a variety of challenging fine‑tuning tasks show that TaRA consistently outperforms existing state‑of‑the‑art initialization methods, offering a simple, robust, and scalable solution for effective LoRA initialization.
By Taehyeon Kim, Eunhyeok Park
VideoPulse is a neonatal dataset and end‑to‑end pipeline that estimates heart rate and peripheral capillary oxygen saturation (SpO2) from facial video without contact. The dataset contains 157 recordings from 52 neonates, and the pipeline uses face alignment, artifact‑aware supervision, and 3D CNN backbones to produce predictions every 2 seconds. On the NBHR dataset the model achieves a heart‑rate MAE of 2.97 bpm and SpO2 MAE of 1.69 %.
"whyItMatters":"The results show that short, unaligned neonatal video segments can accurately estimate vital signs, offering a low‑cost, non‑invasive monitoring option for neonatal intensive care."
By Deependra Dewagiri, Kamesh Anuradha, Pabadhi Liyanage, Helitha Kulatunga, Pamuditha Somarathne, Udaya S. K. P. Miriya Thanthrige, Nishani Lucas, Anusha Withana, Joshua P. Kulasingham
GDB-Reward is a framework that converts diverse graphic design evaluation metrics into a single reinforcement learning reward, enabling prompt optimization for frozen image generators. By using these metrics as scalar rewards, the method improves adherence to design specifications in perceptual quality, rendering fidelity, and spatial accuracy without fine‑tuning the generator. Experiments show that non‑differentiable metrics can serve as effective optimization objectives in settings lacking differentiable supervision.
By Adrienne Deganutti, Purvanshi Mehta, Simon Hadfield, Andrew Gilbert
IDEA is a training‑free, input‑dependent steering method for large language models that matches activations to cluster‑specific directions aligned with a target concept. It clusters positive and negative activation supports per attention head, solves an optimal‑matching problem to create a pool of cluster‑conditional directions, and selects the best match for each input at inference time. This approach preserves the input’s original representation while improving the truth × info rate on TruthfulQA by an average of 9.9% (up to 23.5%) over input‑independent baselines.
By Zheng Wang, Muchen Li, Renjie Liao, Yan Leng
The paper introduces SEAL, a training-time, parameter‑efficient defense that attaches a plug‑and‑play adapter to the shared expert component of Mixture‑of‑Experts models, and SEAL++, which adds an orthogonal constraint to preserve existing safety subspaces. By leveraging the always‑activated shared expert, SEAL mitigates the structural vulnerability of sparse routing to adversarial manipulation, reducing attack success rates by up to 60% with minimal impact on model capability. The approach is evaluated across six attack scenarios involving harmful prompting, jailbreaks, malicious fine‑tuning, and neuron pruning.
By Qingyu Meng, Yiwei Zha, Jiahuan Pei, Koen Hindriks, Herbert Bos, Min Chen
The paper introduces a zero‑shot subject‑driven video generation framework that eliminates the need for per‑subject tuning and large subject‑video datasets. It achieves this by separating identity injection—learned from subject‑image pairs—and motion‑awareness preservation—maintained with a small set of arbitrary videos, and optimizes both with stochastic switching and dropout techniques. Using CogVideoX‑5B, the method adapts a single model with only 200K subject‑image pairs and 4,000 arbitrary videos in 288 A100 GPU hours, representing roughly 1% of the compute required by previous zero‑shot baselines while preserving subject fidelity and motion quality.
By Daneul Kim, Jingxu Zhang, Wonjoon Jin, Sunghyun Cho, Qi Dai, Jaesik Park, Chong Luo
The paper introduces Evidence Sufficiency Boundary Training, a framework that teaches models to abstain from answering until the supplied evidence is fully sufficient, and to remain stable when additional redundant evidence is added. By constructing ordered evidence chains from datasets such as HotpotQA, 2WikiMultiHopQA, and MuSiQue, the method applies level supervision, a boundary flip margin, post‑boundary stability, and answer recall protection. Experiments with Qwen2.5‑3B‑Instruct and LoRA adaptation show improved boundary localization (flip accuracy 0.807 vs 0.781) and a lower unsupported‑answer rate (0.095 vs 0.101) while maintaining competitive raw QA F1.
By Haruto Sato, Yuki Tanaka, Ren Nakamura, Aoi Kobayashi, Mei Ito
DeepAffinity is a model designed to predict eCommerce users’ future preferences for product aspects such as brand, size, and color, treating this as a temporal prediction problem. It uses small language models with structured prompts and specialized prediction heads, outperforming standard generative fine‑tuning and general‑purpose open‑source LLMs that lack task‑specific tuning. The approach improves recommendation quality on a large multinational eCommerce platform.
By Yotam Eshel, Guy Hadad, Guy Feigenblat, Yuri M. Brovman, Matt Gearhart, Bracha Shapira
The paper introduces Tool Primitives, a design that replaces rigid API schemas with natural language interfaces for tool calling, enabling seamless inter-tool communication. It builds ToolFace, a repository of over 25,000 functions that LLMs can dynamically retrieve, and HEART, a harness engineering framework that orchestrates tool use with planning, routing, and verification. Experiments show HEART outperforms fine‑tuned models and leading commercial LLMs while cutting API costs by up to 85%.
By Haibo Jin, Suijin Wang, Xucheng Yu, Haojing Luo, Haohan Wang
hLLM (Hungarian LLM) is a decoding strategy that efficiently produces ranked lists from large language models by decoding all ordinal positions in a single forward pass. It extracts an item‑position score matrix from the model’s hidden states, then applies the Hungarian algorithm to obtain a valid permutation, avoiding the sequential token generation of traditional autoregressive decoding. Experiments show that LoRA‑based fine‑tuning with teacher ranking distillation achieves 28 ms end‑to‑end inference, a 64× speed‑up while preserving ranking quality comparable to the teacher model.
By Emil Laftchiev, Prachi Agrawal, Moe Kayali, Bixing Yan, Qi Xu, Zijie Lei, Chen Qiu, Zhi Hua, Ke Li, Luke Simon