News & Events

Position:Home > News & Events > EECS Spotlight

Artificial Intelligence Foundations

Release time:2017-10-08


Research in artificial intelligence has been a major focus in the area of computer science in recent years. Construct theoretical framework and the scope of artificial intelligence is critical for in-depth understanding of the science of artificial intelligence. In addition, active learning, which aims at reducing the manual annotations required for learning, becomes one of new paradigms of machine learning.

1) Margin theory of Boosting. In the past years, research teams from Peking University solve the problem of understanding why the ability of Adaboost to prevent overfit seems contrary to Occam's razor principle. Equilibrium margin theory was proposed to give a periodical explanation to long-term debate over the Margin theory of Boosting from academic. Research results on above margin theory of boosting and machine learn mechanisms were published in machine learn top journal JMLR and top conferences COLT, NIPS, ICML and IJCAI. The first paper on the margin theory of AdaBoost published in COLT in 2008 is the first COLT paper by authors from mainland China since COLT was founded in 1988.

2) New paradigms of machine learning. PKU researchers proved that label complexity of active learning is less than the passive learning in exponential order in some smoothness conditions, which has reached the international forefront of the current theoretical research. Privacy protection is a prerequisite for learning from sensitive data. On the basis of the concept of differential privacy, novel mechanisms both for query answering and outputting synthetic database in terms of smooth queries was proposed. The new mechanisms outperform all existing methods.

3) Computational Intelligence. Peking University has been leading the research of Computational Intelligence and proposed one of the notable achievements is the fireworks algorithm (FWA), which is a swarm intelligence algorithm in 2010. FWA converges to the optimum much faster than particle swarm optimization. It has also been applied to a wide range of real-world problems, such as multi-objective optimization, nonnegative matrix factorization, swarm robotics, spam detection, document clustering, and inversion of regional seismic waveform. Based on FWA, an efficient search strategy, called GES (Group Explosion Strategy), is proposed to solve the multi-target search problem in swarm robotics. The results of Computational Intelligence have been widely published on premier conferences including AAAI, IJCAI, WCCI, CEC, ICSI, IJCNN and high-impact journals and 5 monographs have been published. Two special issues on fireworks algorithm has been published by International Journal of Swarm Intelligence Research. These achievements have been awarded as “Wu Wenjun artificial intelligence science and technology award” of CAAI (Chinese Association for Artificial Intelligence) in 2016.