When we apply machine learning in real applications, we need to address some important challenges. First, the world is an uncertain place because of physical randomness, incomplete knowledge, noise, ambiguities, and contradictions. It is critical to model uncertainty and draw inference for intelligent systems. Second, ML algorithms (e.g., deep networks) can be vulnerable to some adversarial noise. This is of high risk in high-stakes and security-critical applications. In this talk, I will present some advances in probabilistic machine learning (particularly a ZhuSuan probabilistic programming library and some scalable algorithms) and adversarial attack and defense for deep networks. Some application cases in semi-supervised learning, few-shot learning and crowdsourcing will be highlighted. For adversarial attack and defense, our methods won the first places in all three tasks in NIPS 2017 competition organized by Google Brain. 

机器学习在解决实际应用问题时需要解决一些重要挑战。首先，由于物理随机性、不完全信息、噪声、歧义、冲突等因素，我们要处理的对象普遍存在不确定性。因此，智能系统需要对不确定性进行有效的建模和推理。其次，在对抗噪声的情况下，很多机器学习算法（如深度神经网络）往往比较脆弱，容易被误导，这给高风险、安全敏感的应用带来了很多潜在威胁。在这个报告中，我将介绍概率机器学习的一些进展（特别是珠算概率编程库和一些可扩展的推理算法）以及深度神经网络的对抗攻击与防御，并且介绍一些典型的应用案例，包括半监督学习、小样本学习、众包学习。在对抗攻击与防御方面，我们的部分工作获得谷歌大脑在NIPS 2017组织的国际比赛的所有三个任务的冠军。

演讲提纲：

1. 概率机器学习介绍

2. 珠算概率机器学习编程框架

3. 概率机器学习应用案例

4. 深度神经网络的对抗攻击与防御

听众受益点：

1. 了解概率机器学习的概念和方法

2. 了解概率机器学习的应用实例和编程库

3. 了解深度学习对抗攻击与防御的方法和经验