人类大脑天生就有一种叫做连续学习（Continual Learning）的能力，然而目前的主流神经网络存在一种叫做灾难性遗忘（Catastrophic Forgetting）的问题。所谓灾难性遗忘，即突然和充分地学习一套新的模式，会导致完全遗忘已经学习过的模式。这篇文章提出一种将学习系统看作一个纤维丛（Fiber Bundle）的通用学习模型。通过在一系列的机器学习的实验中将文章提出的模型与传统模型的学习能力进行对比，作者发现文章提出的模型不仅仅有着极佳的连续学习能力，而且具有很大的信息容量。此外，作者发现，在某些学习情景下，假如我们让模型能够对时间产生意识（类似于人类大脑的情景记忆机制），模型的连续学习能力能够进一步得到增强。最后，作者发现文章提出的模型的遗忘性质和人类记忆的遗忘性质有着很好的对应。这个工作或许能够增进我们对人类大脑学习机制的认识。
本文用简单的 pytorch 代码实现一个生成和识别简单 sine 波的生成对抗网络（Generative Adversarial Networks，简称GAN） 。为了便于理解，我尽可能省去了不必要的代码。在阅读本文之前，读者需要对 pytorch 有一定的了解。读者读完这个简单例子，可以学会如何自己构建一个简单的GAN。
This article introduces an automated English-writing reviser (AI-Reviser) based on advanced AI/ML technology. It was developed during the summer of 2017 using my spare time, and has gradually gained popularity (reached more than 78,000 page views in around 6 months) especially among students whose native language is not English. Table of Contents: Motivation Product design … Continue reading AI-Reviser: An automated writing reviser based on cutting-edge artificial intelligence technology.
Nature is a master in coloring. It often minimizes the variety of colors used and arouses comfortable and astonishing sensations. For fun, I wrote a small piece of code (available here) that recolors an image using the coloring style of another image. The current version is quite naive (based on clustering of colors using a … Continue reading Recoloring an image using the coloring style of another image
A power-law distribution for the connectivities of the nodes in a complex network has been widely reported in the studies of diverse networks. Though preferential attachment (PA) has been well accepted as an explanation for the power-law distributions observed in many systems, it is unclear how the asymmetry of information-flow in the networks will influence … Continue reading Documentation for the project “The impact of asymmetric information-flow on the power-law exponent”
The `out of the blue' nature of recent terror attacks and the diversity of apparent motives, highlight the importance of understanding the online trajectories that individuals follow prior to developing high levels of extremist support. Here we show that the physics of stochastic walks, with and without temporal correlation, provides a unifying description of these … Continue reading Documentation for the project “Universality and correlations in individuals wandering through an online extremist space”
The dynamics of online human activity behind the evolution of online groups that individuals follow prior to developing high levels of extremist outbreak is a hot topic and crucial for improving the social security. Here is the documentation for the codes used for the analyses. Table of Contents Installation Preprocessing the datasets (optional) Observations Evolutionary … Continue reading Documentation for the project “Group dynamics”