2026.11.19 | 并行决策生成图像;双路径适配天气预测

2026.11.19 | 并行决策生成图像;双路径适配天气预测

10分钟 ·
播放数30
·
评论数0

⏱️ Highlights
00:27 🎨 Jeveration: System One Models Now Generate(Jeveration:系统一模型如今也能生成)
01:02 🌦️ PROLA: Principal-Orthogonal Low-rank Adaptation for Predictive Spatiotemporal Weather Downscaling(PROLA:用于预测式时空气象降尺度的主方向—正交低秩适配)
01:38 🎮 Game2Player: Scaling Gameplay Experience for Visual Agents through Environment-Player Co-Development(Game2Player:通过环境—玩家协同开发扩展视觉智能体的游戏经验)
02:13 👀 ALIVE: Making Streaming Video Models Interactive(ALIVE:让流式视频模型具备交互能力)
02:49 🧠 CurvDelta: Low-Curvature Delta Modeling For Surgical Learning and Unlearning(CurvDelta:用于精准学习与遗忘的低曲率增量建模)
03:25 🔍 The Resolution Hypothesis: Agent Evolution's Key Is Identifying Improvements(分辨率假说:智能体演化的关键是识别改进)
04:02 🦾 HINT-WAM: Hindsight-Informed Training for World Action Models(HINT-WAM:面向世界动作模型的后见信息训练)
04:38 🌍 WorldRuler: Interactive World Modeling under Configurable Rules(WorldRuler:可配置规则下的交互式世界建模)
05:15 📊 LightTAR: Light and Fast Target-Aware Representations for Multimodal Tabular Learning(LightTAR:面向多模态表格学习的轻量快速目标感知表征)
05:51 🎬 Code World Model 2.0: Towards Infinite Controllable Video Generation(代码世界模型 2.0:迈向无限可控视频生成)
06:27 🖼️ Decoding Without a Decoder: Direct Pixel Prediction from Diffusion Transformers(无需解码器的解码:从扩散 Transformer 直接预测像素)
07:01 ⚙️ Mechanistic Critics for Sample-Efficient NPU Design-Space Exploration(面向样本高效 NPU 设计空间探索的机理评判器)
07:39 💻 Tideway: Phase-Aware Tiered Memory Management for MoE Inference on Consumer Devices(Tideway:面向消费级设备 MoE 推理的阶段感知分层内存管理)
08:14 🛟 PReMind: When Personal Memory Meets Visual Hazards(PReMind:当个人记忆遇上视觉危险)
08:50 ✂️ RoughCut-Bench: A Benchmark for Automatic Video Rough-Cut with Large Vision-Language Models(RoughCut-Bench:面向大型视觉语言模型自动视频粗剪的基准)

播客论文目录配图

【关注我们】
您还可以在以下平台找到我们,获得播客内容以外更多信息
小红书: AI速递