AD-TECH
Lab BLOG
RCO アドテクLab ブログ

AAAI Reports (1st day)

2017/02/06stakaya

このエントリーをはてなブックマークに追加

I am attending Association for the Advancement of Artificial Intelligence (AAAI) at San Francisco. The program is available on this page. I am reporting the brief summary of the sessions everyday. The detail discussion about each research may be written later.

AAAI-17 Invited Talk

The first session was presented by Rosalind Picard at MIT Media Lab(http://web.media.mit.edu/~picard/).

The introduction was started with the story about superiority of the ML algorithm in simple emotion estimations. He developed the application and the device for the estimation. (He has his own company!) In addition, he made the SDK for realtime emotion estimation available online as an open license program.

I enjoy this talk since I am interested in measuring my health data such as heart rate or number of steps. my health data is available here I aim to enhance my daily life or even my entire life by monitoring my health condition.

MLA1: Recommender Systems

I chose to see this session because recommendation system is very familiar in my job. When it comes to matrix decomposition, he mentioned the idea of making use of global structure and the local structure separately for better performance.

He also mentioned the idea about the introduction of the true distribution to evaluating the risk (loss function) and the idea that the social network structure in SNS helps to resolve the problem so called “cold start” by formulating the problem as a lower rank estimation. Various kinds of methods to tackle “cold start” have been proposed, so I learned a lot from this talk.

VIS2: Categorization

This session was about the video processing (mainly classification). The following topics were introduced.

  • Recurrent Neural Network(RNN) based model can extract the “scene” with the help of context.
  • build the state-space model with hidden layer and calculate the similarity between persons to detect the collective movement of humans.

I enjoyed this talk since I am interested in the state-space model but never realized it can be applied to video processing in such ways.

MLA3: Machine Learning Applications

The following topics were introduced.

  • prediction of customer volume by combining the customer action data and the latent customer representation.
  • evaluation of the risk of disease by matrix decomposition
  • prediction of interesting event (such as goal scene in soccer) from the streaming video data.

I realized that the ML methods have been used in various kinds of applications.

I reported on the first day of AAAI!