1 of 11

My Research

Pangfeng Liu

Parallel and distributed computing lab, CSIE, NTU

2 of 11

Summary�

Deep learning neural networks system optimization
AI compiler for deep learning
Heterogeneous system architecture for deep learning
Resource scheduling
Algorithm design and analysis

3 of 11

My Research

My research in the last five years has been focusing on the parallel and distributed computing, including deep learning system optimization, job scheduling, machine learning on GPU, databases, systems for big data, and cloud computing.

4 of 11

Deep Learning System Optimization

In recent years, deep neural networks have achieved remarkable success in several domains of computer vision.
Deep neural networks outperform traditional computer vision approaches. However, the high computation complexity and a large number of parameters make it extremely difficult to train such networks on resource limited environments or applications that require real-time response.
Due to the increasing number of parameters and the limitation of hardware/software resources, model compression has become an important issue.

5 of 11

Deep Learning System Optimization

Parameter pruning
Parameter quantization
Pipeline execution
Parallel execution on multiple GPUs
Checkpointing execution
You need to talk to me or read my papers to know the details.

6 of 11

Scheduling for Parallel and Distributed Systems

Distributed computing is an important mechanism for utilizing computing resources that are distributed in different locations but organized into an integrated service.
In particular, a distributed system provides computing resources that enable users in different locations to utilize the CPU cycles of remote sites. In addition, users can access important data that are only available in certain locations, without the overheads of replicating them locally.
These services are provided by an integrated service platform, which helps user access resources easily and effectively.

7 of 11

Algorithm and Data Structures

Identify system bottleneck and develop new algorithms to improve computer system performance.
Also require implementation and tuning
A recent example is DenseNet scheduling, a result from a special project student.

8 of 11

Skill set

Algorithm design and analysis
Computer system knowledge
Programming
English technical writing�

9 of 11

Funding

Ministry of Science and Technology, Taiwan
Computer system lab, Institute of Information Science, Academia Sinica

10 of 11

Contact

Rm 425, NTU CSIE building
pangfeng@csie.ntu.edu.tw
Google "Pangfeng Liu"