首页 博客

2019年最受关注的人工智能影响者

Yoshua Bengio: Yoshua BengioOCFRSC (born 1964 in Paris, 法国)是加拿大人 计算机科学家,因在 人工神经网络 and deep learning. 他是2018年的共同获奖者 ACM A.M. Turing Award for his work in 深度学习。 He is a professor at the Department of Computer Science 和 Operations Research at the 蒙特利尔大学 和...的科学总监

研究人员检查三种内在动机类型,以激发强化学习(RL)代理的内在目标

强化学习(RL)使工具能够直接从高角度图像输入(例如运动,机器人操纵和游戏)中决策并解决未知环境中的复杂问题。但是,这些成功是建立在对手工制作的奖励功能进行深入监督的基础上的。代理商会根据他们的表现受到奖励和惩罚,并最终获得奖励。

脸书 AI推出了DeiT(数据高效型图像变压器):一种训练计算机视觉模型的新技术

Facebook AI开发了一种称为 数据高效的图像变压器(DeiT) 训练利用变压器的计算机视觉模型来解锁人工智能许多领域的重大进步。  DeiT requires far fewer data 和 far fewer computing resources to produce a high-performance image classification model. In training a DeiT model with just a single...

斯坦福大学研究人员推出LUCIDGames,这是一种可以预测和规划自动驾驶汽车自适应轨迹的计算技术

斯坦福大学的研究人员最近推出了LUCIDGames, 计算技术  to predict 和 plan adaptive trajectories for autonomous vehicles. This technique integrates an algorithm based on game theory 和 an estimation method. 人们通常可以找出其他驱动因素' goals in their surroundings 和 negotiate decisions, for example, who goes first at a given intersection. In...

Understanding 的 Memorization Of Data Including Personal Identifiable Information in GPT-2 Model

The 伯克利人工智能研究 (BAIR)评估了多大的语言模型 memorize and regurgitate their training data's rare snippets in a recent paper. The focus was on GPT-2 和 found that at least 0.1% of its text generations contain lengthy verbatim strings, "copy-pasted"从其培训集中的文档中获取。 对于在...上训练的语言模型,这样的记忆将是一个突出的问题。

一家印度初创公司正在使用AI通过电话传感器来绘制印度的坑洼

India is the seventh-largest country globally 和 has one of the world's largest road networks, with 59 lakh km of road length. Potholes in India have become so common that drivers have learned to spot them 和 violently avoid them. However, this causes further accidents. Pothole death has been increasing because of heavy traffic 和 water on...

这个大学交流平台,‘InSpace’,使用TensorFlow.js作为聊天中的毒性过滤器

InSpace is a virtual communication 和 learning platform. It helps people interact, collaborate, 和 educate in familiar physical ways, but in a virtual world. It is designed to experience the fluid, personal, 和 interactive nature of a real classroom. It helps participants break free of “Brady Bunch” boxes in existing conference solutions to create a fun, natural, 和 engaging environment...

使用CNN和变压器将感应图像偏置编码为模型的新方法

Researchers at Heidelberg University have recently proposed a novel method to efficiently code inductive image biases into models while retaining all transformers’ flexibility. This approach combines the inductive bias’s effectiveness in convolutional neural networks (CNNs) with transformers’ expressivity to model 和 synthesize high-resolution images. 变形金刚的局限性 变压器已显示出令人鼓舞的结果...

深心 Introduces MuZero That Achieves Superhuman 性能 In Tasks Without Learning 的ir Underlying Dynamics

以前,DeepMind使用强化学习来教程序以掌握各种游戏,例如中文棋盘游戏'Go,'日本战略游戏'Shogi,' chess, 和 challenging Atari video games, where earlier AI programs were taught the rules first during training. DeepMind引入了MuZero,该算法(通过结合基于树的搜索...

与McAfee高级数据科学家Sherin Mathews的独家谈话

Asif: Tell us about your journey in AI 和 machine learning so far.  What factors influenced your decision to pursue a PhD 和 a career in the field of AI?  Sherin: I was initially intrigued by the field of Machine Learning (ML) 和 Deep Learning (DL) as it presented a world of endless possibilities...

深心协调了现有的神经网络限制以胜过神经符号模型

Neural networks have achieved success in various perceptual tasks. However, it is stated that they are ineffective in solving problems requiring higher-level reasoning. Recent experiments with two recently released video question-answering datasets (CLEVRER 和 CATER) show that neural networks cannot adequately reason about the Spatio-temporal 和 compositional structure of visual scenes. On the other...
Join the AI conversation 和 receive newsletters, offers & invitations.