Haiping Huang

Associate Professor

Email: physhuang [AT] gmail [DOT] com; huanghp7[AT]mail.sysu.edu.cn

Research Interests

• Our theoretical work on statistical properties of deep networks under weight removal perturbations is updated on arXiv: 1705.00850. The new version clarifies a first-order phase transition separating a paramagnetic phase from a spin glass phase, which may connect to robustness of the deep networks and performance of the dropconnect algorithm when different dropconnect probabilities are used. To be published by Phys Rev E (2018)
  
• A theory about deep computation was just posted on arXiv: 1710.01467.
  
• I was invited to give a talk in statistical physics subfield of annual conference of Chinese physics society (Sept 8-10, Si-Chuang University).
• Our new work on redundancy of active paths in deep networks is on arXiv: 1705.00850. Abstract accepted as a poster in SNL-2017
  
• Interesting computational role of zero synapses was just reported in recent work arXiv: 1703.07943 : Our mean-field theory predicts surprising computational role of these synapses. A rapid decrease of their number signals concept formation in a neural network, and zero synapses refine the learned concept by serving as contour detector. Zero synapses are also sensitive to the contrast level of sensory inputs. This paper will be published in J. Phys. A as a letter (intended for high quality research, reporting significant new results)
• I will give a talk in a wonderful conference "machine learning and many-body physics' at Kavli Institute for Theoretical Science (July, Beijing).
• I was just awarded 8th RIKEN Research Incentive award (to young RIKEN scientists for their significant scientific achievements). News Report
  
• Reviewer thank you from PLoS Computational Biology (2016)
  
• A new algorithm to improve greatly standard back propagation in deep network training was announced on arXiv:. 1701.07974.
• Statistical mechanics of unsupervised feature learning in a restricted Boltzmann machine with binary synapses  arXiv:1612.01717, in quora, accepted by J. Stat. Mech (2017),  talk at Prof. Nishimori 's group on Jan 18, 2017
  
• Recent work on unsupervised feature learning based on the restricted Boltzmann machine learning, our theory shows that the machine can only understand the feature of the data by showing it sufficient data.
• Recent work on theory of population coupling appears on arXiv (2016). presented at CNS2016 meeting as a poster. accepted by J. Stat. Mech (2017).
  
• Recent work on Clustering of neural codewords revealed by a first-order phase transition appears on arXiv (2016). twitter selected by MIT Technology Review contributed talk at a workshop of CNS2016

• 2018.05–present Associate Professor, Sun Yat-sen University, China, Lab for statistical physics and neural computation.
• 2014.08–2018.05 Research Scientist, RIKEN Brain Science Institute, Saitama, Japan.
  Lab for neural computation and adaptation, in collaboration with Taro Toyoizumi.
• 2012.08–2014.08 JSPS Postdoctoral Fellow, Department of Computational Intelligence and Systems
• Science, Tokyo Institute of Technology, Yokohama, Japan.
  Statistical mechanical approach to massive probabilistic inference, in collaboration with
  Yoshiyuki Kabashima.
• 2011.08–2012.08 Visiting Scholar, Department of Physics, The Hong Kong University of Science and
  Technology, Clear Water Bay, Hong Kong.
  Algorithmic design and analysis of random optimization problems, in collaboration with K Y Michael Wong.
• 2006.09-2011.07 PhD, Theoretical Physics, Institute of Theoretical Physics, Chinese Academy of Sciences, Beijing, China. Applications of Statistical Physics to Graphical Models of Statistical Inference, supervised by Haijun Zhou.
• 2002.09-2006.06 Bachelor of Science, Physics, Sun Yat-sen University, Guangzhou, China.

Honors and awards

Mar 2017  8th RIKEN Research Incentive Award, RIKEN

Jan 2012 JSPS Postdoctoral Fellowship for Foreign Researchers, Japan Society for the Promotion of Science (JSPS)

Referee Services
Physical Review Letters (9), Physical Review E (10), Communication in Theoretical Physics (2), Journal of Physics: Conference Series (2), Journal of Statistical Mechanics: Theory and Experiment (7),  Journal of Physics A: Mathematical and Theoretical (1), Eur. Phys. J. B (1), Neural Networks (2), Physical Review X (2), Scientific Reports (1), Network Neuroscience (1), Neurocomputing (1), PLoS Comput Biol (2), Physica A (1)