Tag
RTL-Interact is a multi-task post-training framework that trains a single RTL-specialized LLM to generate RTL, construct EDA checks, and revise designs using tool feedback, paired with Pattern-Aware Policy Optimization (PAPO) for multi-task RL. On CVDP, RTL-Interact-32B reaches 48.7% weighted Pass@1 and 68.9% Pass@5 (surpassing DeepSeek-V3.2), while PAPO cuts average tool calls by 63.8% and the models outperform RTL-specialized baselines on VerilogEval v2 and RTLLM v2.
The paper proposes AICG, an AI-aware cognitive-supervised two-branch framework for automated mathematical short-answer grading that combines RoBERTa-based scoring with Q-matrix skill prediction and a Gen-AI detector to weight unreliable responses, reducing MAE by 50.3% and improving QWK from 0.798 to 0.927 over RoBERTa regression.
The paper proposes MOSAIC, a modality-separated architecture with inter-modal cross-attention and modality-partitioned gradient routing, which improves average macro-F1 by 7.4 points across four tasks on the CEJC corpus while reducing backbone parameters by 72% versus fusion-based baselines.
CONCUR introduces a new advantage estimator for multi-task, multi-reward RL training of LLMs that projects gated auxiliary rewards onto scores concordant with primary rewards, outperforming baselines across math, code, instruction following, and writing tasks on three models.
This paper proposes Error-Subspace Calibration (ESCal), a post-merging calibration framework that identifies low-dimensional error subspaces from weight differences between expert and merged models, using lightweight Bridge modules to recover task-specific behavior. Experiments on vision and NLU benchmarks show ESCal improves merged-model performance across diverse architectures and merging methods.
WeVisOne-VL (WeChat Vision All-in-One) 是一个统一对象级与像素级视觉感知的视觉语言模型，通过结构化 token 预测与轻量像素解码器支持检测、定位、计数、OCR、分割和抠像等任务，在30多个基准上取得领先性能，并已开源 2B/4B 模型。
The paper proposes EA-SURF, an a priori method that adaptively tunes scalarization weights during training so multi-task speech LLMs converge toward user-specified task preferences along the Pareto front, improving normalized hypervolume by 17.8% over fixed loss weighting across ASR, timestamp prediction, and speech translation tasks.
VX-One 是一个小型多任务神经网络，输出带类型且经过温度校准的决策基元（校准门控、含显式未知选项的闭集选择、序数分数），用于实时边界流量分诊。论文在 UNSW-NB15、CIC-IDS2017、EMBER2024 上评测其与签名引擎的两阶段级联表现，发现其在时漂移下检测占优，但校准会崩溃且未知攻击拒识能力跨基准迁移性差。
研究人员提出BioBigBird，一个基于稀疏注意力机制、可处理最长4096 token的双向生物医学预训练语言模型，并通过多任务学习联合优化NER和关系抽取，在BLURB基准上取得具有竞争力的结果。
论文提出 JIVEAdapter，一种受统计学 Joint and Individual Variation Explained 启发的多任务加性低秩适配器，将权重更新分解为跨任务共享的 Joint 结构与任务专属且近正交的 Individual 结构，并在 GLUE/SuperGLUE 上与 DeBERTaV3-base 取得竞争力表现。
本文提出 Hebero，一个基于 GPU 并行 Isaac Lab 的异构多任务机器人操作基准，以及演示引导策略优化方法 DGPO（含 IW-ABC），在 40 项任务上以 50 条演示/任务达到 90.1% 状态输入平均成功率，超越 FAMO-ABC 7.8 个百分点，并在真实 Piper 机器人上验证了单一多任务策略的迁移能力。
Maxime Labonne highlights a single model capable of playing multiple games in parallel, suggesting AI advances that go well beyond text generation for the video game industry.
论文提出 Adaptive Mutual Distillation (AMD) 框架，通过两个采用不同任务平衡策略的模型进行协同蒸馏，并利用短训练探针和任务级验证分数自适应调整各任务与传输方向的蒸馏权重，在六个基准和三个 LLM 骨干上取得优于 SFT 与现有任务平衡方法的效果，合并后平均提升 2.91 分。
提出一种自进化 agent harness 框架：同一冻结模型先作为 solver 解题、再作为 proposer 直接编辑自己的 harness 代码，在多任务上进化后于分布外基准上显著超越 Codex（提升 12.64 分）并达成匹配表现。
This paper proposes EN-HMTFL, an encoder-sharing hierarchical multi-task federated learning framework for vehicular ad hoc networks that lets vehicles train heterogeneous perception tasks collaboratively via a shared encoder while keeping task-specific decoders local, improving accuracy by up to 24% and reducing communication rounds by up to 28.8%.
The paper investigates how visual generation supervision can improve visual understanding capabilities, introducing OmniTaskonomy as a unified taxonomy to analyze task-dependent benefits and guide training curricula.
The paper proposes Marathoner, an autonomous agentic model for ultra-long-horizon execution, using a comprehensive post-training pipeline with tasks synthesized from GitHub PRs and a novel reward strategy to achieve superior performance on benchmarks.
DuoOPD 论文提出一种新的多任务在线策略蒸馏方法，利用教师-学生的联合结果来决定反馈方向与教师支持方式，在 Qwen3 和 Llama 上相比标准 OPD 平均宏准确率提升 2.58 和 5.98 个百分点。
This paper proposes a framework to learn cross-task relationships in multi-task models by approximating joint label distributions, improving performance in YouTube's recommendation systems through transfer learning.
AI capabilities for expression recognition, object detection, and finger counting have achieved less than 1 second latency, highlighting notable real-time performance gains.