Modern processor design and research in both industry and academia rely on early-stage modeling and simulation. The ideas that make up every CPU, GPU, and accelerator you have ever used started their life in cycle-level C++ processor simulation. Today, much of the progress we see in the processor industry comes from specialization (i.e. Google’s TPU) and acceleration (i.e. GPGPU computing, such as NVIDIA’s CUDA). The Accelerator Architecture Lab at Purdue (AALP) develops a popular open-source GPU simulator called Accel-Sim that models modern NVIDIA GPUs executing compute workloads, like those commonly used in machine learning. Intimate details of the actions taken on each cycle of a real processor are modeled in Accel-Sim’s C++ code, such that new architectural ideas can be explored and empirically evaluated on real workloads. Simulating such an advanced, scaled system consumes a significant amount of CPU-time and memory (for some workloads - on the order of a TB!). This summer project involves understanding the high-level design of GPUs, the basic operation of CUDA, and optimizing the simulator infrastructure to consume orders of magnitude less memory at runtime, enabling larger and more complex workloads to be simulated. The successful completion of the project will see the student contribute to a highly-visible piece of open-source software and develop foundational skills to work at hardware design companies like Intel, AMD, NVIDIA, Qualcomm, Microsoft, and many others.

More information: https://accel-sim.github.io
Group Website: https://engineering.purdue.edu/tgrogers/group/aalp.html

Big Data/Machine Learning, Deep Learning, Other

Computer Engineering, Computer Science

C/C++ and Python experience Knowledge of computer architecture a plus

ECE

Timothy Rogers

Modern processor design and research in both industry and academia rely on early-stage modeling and simulation. The ideas that make up every CPU, GPU, and accelerator you have ever used started their life in cycle-level C++ processor simulation. Today, much of the progress we see in the processor industry comes from specialization (i.e. Google’s TPU) and acceleration (i.e. GPGPU computing, such as NVIDIA’s CUDA). The Accelerator Architecture Lab at Purdue (AALP) develops a popular open-source GPU simulator called Accel-Sim that models modern NVIDIA GPUs executing compute workloads, like those commonly used in machine learning. Intimate details of the actions taken on each cycle of a real processor are modeled in Accel-Sim’s C++ code, such that new architectural ideas can be explored and empirically evaluated on real workloads. Although the basic design of GPUs share many similarities across generations and vendors, each part and company have subtle differences that can greatly affect their performance on critical applications, such as those found in machine learning. This summer project involves understanding the high-level design of GPUs, the basic operation of CUDA, and modeling the performance of bleeding-edge GPU parts from both NVIDIA (an Ampere A100) and AMD. The successful completion of the project will see the student contribute to a highly-visible piece of open-source software and develop foundational skills to work at hardware design companies like Intel, AMD, NVIDIA, Qualcomm, Microsoft, and many others.

More information: https://accel-sim.github.io
Group Website: https://engineering.purdue.edu/tgrogers/group/aalp.html

Big Data/Machine Learning, Deep Learning, Other

Computer Engineering, Computer Science

C/C++. Computer Architecture and Digital Design Background. Students should be comfortable reading assembly language (ECE 362 equivalent)

ECE

Tim Rogers

Teamwork is an essential competency highly valued by both academia and industry, especially for engineers who usually work in a small group. With tens of years' development, our research group, the Comprehensive Assessment of Team Member Effectiveness (CATME), had collected millions of survey data, including peer comments. The selected SURF student will join our research group to assist with data cleaning, preparation, and analysis for educational or technical research related to teamwork, and perhaps NLP (NLP is not necessary but a plus).

Big Data/Machine Learning, Deep Learning, Learning and Evaluation, Other

ECE, CS, IE, education, social science, management, linguistics, and others

data analysis experience with R, Python, and etc.; familiar with NLP and software programming would be a plus.

Engineering Education

Matthew Ohland

In this project, we will develop cool machine learning approaches to automatically learn bug/vulnerability patterns and fix patterns from historical data to detect and fix software bugs and security vulnerabilities. This project is partially funded by a Facebook Research Award (https://research.fb.com/programs/research-awards/proposals/probability-and-programming-request-for-proposals/).

Earlier work can be found here: https://www.cs.purdue.edu/homes/lintan/publications/deeplearn-tse18.pdf

Big Data/Machine Learning, Cybersecurity, Deep Learning, Other

Compuer Science; Computer Engineering

Good programming skills and strong motivation in research are required. Background in security or machine learning is a plus.

Computer Science

Lin Tan

This project will use public and custom datasets of images/sounds/videos for training deep learning models and explore methods to improve generalization abilities of the trained models.

Deep Learning

Computer Engineering, Computer Science, Electrical Engineering, Computer and Information Technology

Python programming and exposure to ML/deep learning libraries is preferred

Agricultural and Biological Engineering

Dharmendra Saraswat

While intelligent systems promise to extend human capabilities within occupational settings, workers must increasingly collaborate with artificial intelligence (AI) to achieve desired outcomes. This research aims at enhancing bi-directional interaction between workers and robots at the construction jobsites by obtaining continuous neurophysiological and psychophysiological data from workers. The developed personalized AI will measure, adapt, and enhance the skill performance of the next generation of the workforce to work safely and communicate effectively in the future automated jobsites.

Big Data/Machine Learning, Deep Learning, Engineering the Built Environment

All Engineering Fields or Neuroscience

Programming (Python, Matlab, C++), Data Analytics (Machine learning, Deep-learning) Seeking applicants who are creative and passionate to explore new areas

School of Civil engineering and CEM

Sogand Hasanzadeh

Computer programs are difficult to evaluate due to the large number of possibilities. Existing evaluation systems are restricted to simple programs or impose restrictions to limit possibilities. This project aims to build an online assessment system that can evaluate non-trivial programs and assist students learning computer programming.

Big Data/Machine Learning, Cybersecurity, Deep Learning, Learning and Evaluation

computer engineering, computer science, electrical engineering

at least two courses on computer programming

Electrical and Computer Engineering

Yung-Hsiang Lu

We will build cool and novel techniques to make deep learning code such as TensorFlow and PyTorch reliable and secure. We will build it on top of our award winning project (https://www.cs.purdue.edu/homes/lintan/publications/variance-ase20.pdf), which won an ACM SIGSOFT Distinguished Paper Award! There may be opportunities to collaborate with Microsoft. See below for more details.

Machine learning systems including deep learning (DL) systems demand reliability and security. DL systems consist of two key components: (1) models and algorithms that perform complex mathematical calculations, and (2) software that implements the algorithms and models. Here software includes DL infrastructure code (e.g., code that performs core neural network computations) and the application code (e.g., code that loads model weights). Thus, for the entire DL system to be reliable and secure, both the software implementation and models/algorithms must be reliable and secure. If software fails to faithfully implement a model (e.g., due to a bug in the software), the output from the software can be wrong even if the model is correct, and vice versa.

This project aims to use novel approaches including differential testing to detect and localize bugs in DL software (including code and data) to address the testing oracle challenge. Good programming skills and strong motivation in research are required. Background in deep learning and testing is a plus.

Big Data/Machine Learning, Deep Learning, Other

Computer Science; Computer Engineering; Data Science

Computer Science

Lin Tan