SSIST by-talk Day 3

There is an idea appears in my mind… to attend a PHD program in ShanghaiTech

Malik and Girod … big bosses’ talk are always too big for me.

Tieyan Liu

Distributed deep learning

• Data partition
• Local shuffling vs global shuffling
• Sampling != shuffling

Aggregation

• Deep learning process is not convex
• Loss function is convex

Yuandong Tian

ELF (extensive, lightweight and flexible) framework for game research

Xilin Chen

Dual purpose hash

Chenglin Liu

Document image analysis (DIA)

Ming Zhou

NMT (tree LSTM)

• Tree to sequence
• Sequence to tree

Knowledge based semantic embedding

Topic aware IR model

• Topic aware neural generative model
• MLP
• Multi_column CNN
• Real time translation

Hard EM / Softmax EM / CRF autoencoder

Thoughts

Good sleep == concentration on interesting topic

Campus life is wonderful but somehow depressive

Conference notes from SSIST 2017 Day 3.