爱可可 AI 前沿推介(10.7)

文章预览

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 AS - 音频与语音 RO - 机器人 1、[CL] ENTP：Encoder-only Next Token Prediction 2、[CV] Depth Pro：Sharp Monocular Metric Depth in Less Than a Second 3、[CL] A generative framework to bridge data-driven models and scientific theories in language neuroscience 4、[CL] Distilling an End-to-End Voice Assistant Without Instruction Training Data 5、[LG] Thermodynamic Bayesian Inference 摘要：基于encoder-only的下一词预测、不到一秒获得清晰单目度量深度图、连通语言神经科学数据驱动模型和科学理论的生成式框架、无需指令训练数据的端到端语音助手蒸馏、热力学贝叶斯推理 1、[CL] ENTP: Encoder-only Next Token Prediction E Ewer, D Chae, T Zeng, J Kim… [University of Wisconsin-Madison & Korea University] ENTP：基于encoder-only的下一词预测 要点: 对传统观点的挑战：本文挑战了广泛接受的观点，即在decoder-only Transformer中 ………………………………