The student will be studying the basic concepts of programming techniques, as well as Machine Learning and Human-Computer-Interaction techniques. Once the knowledge is well obtained, the student will be involved in a research project working on the topic of 3D reconstruction of Hand and Object interaction, utilizing the skill learned. The final stage of this project will be an academic publication and a detailed report on the topics being discussed over the semester. Three credit hours are to be registered, and the student is expected to work 10-12 hours per week on this project.

Deep Learning, Human Factors

• No Major Restriction

Electrical & Computer Engineering

Alex Quinn

Undergraduate researchers enthusiastic about applying artificial intelligence and machine learning (AI/ML) algorithms for a wide variety of data science and engineering-based tasks within industry including data masking, synthetic data generation, cybersecurity, additive manufacturing, software security, and intrusion detection.

Big Data/Machine Learning, Deep Learning, Internet of Things (IoT)

• No Major Restriction

- Basic programming experience in Python/MATLAB - enthusiasm for practical applications of linear algebra, statistics, and computer science-based application is preferred. - strong background in physics and mathematics

Nuclear Engineering

Hany Abdel-Khalik

The student will work with a group of researchers at Purdue, Harvard and Tuskegee University on an NSF Future Manufacturing project focused on internet of things (IoT) edge devices and artificial intelligence (AI) for manufacturing applications. The IoT devices will be deployed at local manufacturing companies and their data will be used to improve operations.

Big Data/Machine Learning, Deep Learning, Internet of Things (IoT)

• Electrical Engineering
• Mechanical Engineering
• Industrial Engineering
• Computer Science
• Computer Engineering

Electrical and Computer Engineering

Ali Shakouri

This project will use deep learning models to analyze sequences of data (such as music). The analysis results will trigger a generative model to create visual art (image or video). Different styles of music (such as class, jazz, and rock) will be used as the input. The music will have different tempos. The computer models analyzes the style and tempo of the music and sets the parameters to generate the visual art. Faster music produces fast moving video. The SURF student will evaluate the existing (open source) computer models for music analysis and visual art generation, integrate them, and provide proof-of-concept demonstrations.

Big Data/Machine Learning, Deep Learning, Human Factors

• Computer Engineering
• Computer and Information Technology
• Computer Science
• Music
• Data Science

Required: At least one course on computer programming. Desired: Knowledge about machine learning and music.

Electrical and Computer Engineering

Yung-Hsiang Lu

*** Desired experience: Strong coding skills and motivation in research are required. Background in deep learning, security, and natural language processing is not required but a plus.

*** Possible industry involvement: Some of these projects are funded by Meta/Facebook research awards and J.P.Morgan AI research awards. 

*** We especially encourage applications from women, Aboriginal peoples, and other groups underrepresented in computing.

*** Project 1. Data-Free Model Extraction

Many deployed machine learning models such as ChatGPT and Codex are accessible via a pay-per-query system. It is profitable for an adversary to steal these models for either theft or reconnaissance. Recent model-extraction attacks on Machine Learning as a Service (MLaaS) systems have moved towards data-free approaches, showing the feasibility of stealing models trained with difficult-to-access data. However, these attacks are ineffective or limited due to the low accuracy of extracted models and the high number of queries to the models under attack. The high query cost makes such techniques infeasible for online MLaaS systems that charge per query.

In this project, we will design novel approaches to get higher accuracy and
query efficiency than prior data-free model extraction techniques.

Early work and background can be found here: 
https://www.cs.purdue.edu/homes/lintan/publications/disguide-aaai23.pdf

*** Project 2. Language Models for Detecting and Fixing Software Bugs and Vulnerabilities

In this project, we will develop machine learning approaches including code language models to automatically learn bug and vulnerability patterns and fix patterns from historical data to detect and fix software bugs and security vulnerabilities. We will also study and compare general code language models and domain-specific language models.

Early work and background can be found here: 
Impact of Code Language Models on Automated Program Repair. ICSE 2023. Forthcoming.
KNOD: Domain Knowledge Distilled Tree Decoder for Automated Program Repair. ICSE 2023. Forthcoming.
https://www.cs.purdue.edu/homes/lintan/publications/cure-icse21.pdf
https://www.cs.purdue.edu/homes/lintan/publications/deeplearn-tse18.pdf

*** Project 3. Inferring Specifications from Software Text for Finding Bugs and Vulnerabilities

A fundamental challenge of detecting or preventing software bugs and vulnerabilities is to know programmers’ intentions, formally called specifications. If we know the specification of a program (e.g., where a lock is needed, what input a deep learning model expects, etc.), a bug detection tool can check if the code matches the specification. 

Building upon our expertise on being the first to extract specifications from code comments to automatically detect software bugs and bad comments, in this project, we will analyze various new sources of software textual information (such as API documents and StackOverflow Posts) to extract specifications for bug detection. For example, the API documents of deep learning libraries such as TensorFlow and PyTorch contain a lot of input constraint information about tensors. Language models may be explored.

Early work and background can be found here: 
https://www.cs.purdue.edu/homes/lintan/projects.html

*** Project 4. Testing Deep Learning Systems  



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 paper (ACM SIGSOFT Distinguished Paper Award)! 



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. 



Early work and background can be found here: 
https://www.cs.purdue.edu/homes/lintan/publications/eagle-icse22.pdf
https://www.cs.purdue.edu/homes/lintan/publications/fairness-neurips21.pdf
https://www.cs.purdue.edu/homes/lintan/publications/variance-ase20.pdf
https://www.cs.purdue.edu/homes/lintan/publications/cradle-icse19.pdf


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

• Computer Science
• Computer Engineering
• software engineering

https://www.cs.purdue.edu/homes/lintan/

Lin Tan

Currently, most cancer diagnosis procedures include a diagnostic imaging process, such as a CT scan followed by a tumor biopsy. Tissue biopsy is an invasive and painful procedure and may pose health risks for patients such as those with kidney diseases. Liquid biopsy, the ability to detect and monitor disease through biofluids, is highly promising and may replace tissue biopsy with an immense potential public health impact. The use of liquid biopsy offers numerous advantages in the clinical setting, including its non-invasive nature, a suitable sample source for longitudinal disease monitoring, a better screenshot of tumor heterogeneity, and lower costs compared to tissue biopsy. Increasing evidence indicates an important cellular function of exosomes and other extracellular vesicle (EV) particles in tumor biology and metastasis, presenting them as intriguing sources for biomarker discovery and disease diagnosis. However, the vast majority of current exosome/EV studies focus on their miRNAs, with few studies on functional proteins such as phosphorylated proteins. As phosphorylation is a major player in cancer and other disease progression, EV phosphoproteins are expected to become actively pursued targets for in vitro disease diagnosis. We were the first group to demonstrate that many phosphoproteins in exosomes and microvesicles could be extracted from a small volume of biofluids, identified by high-resolution mass spectrometry (MS), and verified as potential cancer markers (Chen et al PNAS 2017). In this project, we will focus on non-invasive cancer detection by coupling CT scans with liquid biopsy to eliminate the need for surgery by more than 50%. The IU Urology team led by kidney surgeon Dr. Boris and Dr. Tao’s lab at Purdue University collaborated with prior funding have established specific biosignatures found in low- and high-grade clear cell RCC. An undergraduate student may be involved in the protein sample preparation from biofluids and tissues, maintenance of equipment, and/or bioinformatics analysis.

Big Data/Machine Learning, Biological Characterization and Imaging, Deep Learning, Medical Science and Technology, Nanotechnology

• Computer Science
• Biochemistry
• Biomedical Engineering
• Chemistry
• Biology

Certain coding skills and biostatistics are highly desirable but not required.

Biochemistry AND Chemistry

W. Andy Tao

We are developing deep learning-based workflows for particle diffusometry. Our algorithms are robust against arbitrary flows and thermal gradients. Even though the algorithms have the potential to make in-lab and in-field testing robust, these have never been tested in the lab or the field. The summer student can do various tasks:
1. Microscopic data acquisition
2. Training neural networks

Deep Learning

• No Major Restriction

Weldon School of Biomedical Engineering

Jacqueline Linnes

Despite being widely used, global end-to-end learning has several key limitations. It requires centralized computation, making it feasible only on a single device or a carefully synchronized cluster. This restricts its use on unreliable or resource-constrained devices, such as commodity hardware clusters or edge computing networks. Localized deep learning has the potential to develop highly decentralized, parallel, asynchronous, and fault-tolerant algorithms that can learn on heterogeneous hardware devices under dynamic conditions while maintaining comparable model performance. The long-term vision would be an "Internet of AI" where devices can continuously learn in any conditions.

REU participants will be part of a collaborative team focused on developing novel localized deep learning approaches. One particular target project is a novel localized deep learning approach that we have named a Minimal Learning Unit (MLU). The goal is to create a learning algorithm with local objectives that learns rich unsupervised representations in a highly decentralized and fault-tolerant way. As one specific context, suppose a sensor network should be trained to detect a complex or global event such as anomalous activity over a large area of the wilderness. Each sensor has a very incomplete picture of the situation and can communicate with nearby sensors but cannot communicate with a global centralized server. The goal is to implement both width-parallel and depth-parallel learning on an unreliable set of sensor devices that have limited compute power. This project will focus on the fundamental aspects of novel local learning mechanisms in this highly decentralized environment.

Big Data/Machine Learning, Deep Learning

• No Major Restriction

ECE

David Inouye

The ability to create 3D structures in the micro and nanoscale is important for many applications including electronics, microfluidics, and tissue engineering. This project deals with development and testing of a setup for building 3D structures using a femtosecond pulsed laser. A method known as two photon polymerization is used to fabricate such structures in which a polymer is exposed to laser and at the point of the exposure the polymer changes its structure. Moving the laser in a predefined path results in the desired shape and the structures. The setup incorporates the steps from designing a CAD model file to slicing the model in layers to generating the motion path of the laser needed for fabricating the structure. Possible improvements to the process by the undergraduate researcher include control algorithms, better CAD models, and better manufacturing strategies.

Deep Learning, Fabrication and Robotics, Material Processing and Characterization, Nanotechnology

• Mechanical Engineering

junior or senior standing, CAD, Matlab or Python, minimum GPA > 3.5

Mechanical Engineering

Xianfan Xu

Atmospheric blocking events and 'Bomb Cyclones' are an important contributor to high impact extreme weather events. Both these weather extremes lead to heat waves, cold spells, droughts, and heavy precipitation episodes, which have dire consequences for the public health, economy, and ecosystem. For example, the blocking-induced heat waves of 2003 in Europe led to tens of thousands of human casualties and tens of billions of dollars of financial damage.

Traditionally, prediction of extreme weather events is based on direct numerical simulation of regional or global atmospheric models, which are expensive to conduct and involve a large number of tunable parameters. However, with the rapid rise of data science and machine learning in recent years, this proposed work will apply convolutional neural network to an idealized atmospheric model to conduct predictability analysis of extreme weather events within this model. With this proposed machine-learning algorithm, our project will provide a robust forecast of heat waves and atmospheric blocking with a lead-time of a few weeks. With more frequent record-breaking heat waves in the future, such a prediction will offer a crucial period of time (a few weeks) for our society to take proper preparedness steps to protect our vulnerable citizens.

This project is based on developing and verifying the machine learning algorithm for detecting extreme weather events in an idealized model. We will use Purdue’s supercomputer Bell to conduct the simulations. The undergraduate student will play an active and important role in running the idealized model, and participate in developing the algorithms. As an important component of climate preparedness, the proposed work aims to develop a physics-informed machine learning framework to improve predictability of extreme weather events.

Closely advised by Prof. Wang, the student will conduct numerical simulations of an idealized and very simple climate model, and use python-based machine learning tools to predict extreme weather events within the model. Prof. Wang will provide weekly tutorial sessions to teach key techniques along with interactive hands-on sessions. The students will get access to the big datasets on Purdue’s Data Depot, analyze and visualize data of an idealized atmospheric model. The student will use convolutional neural networks (CNNs) to train and assess a Machine-Learning model. The student will further use feature tracking algorithm to backward identify the physical structure in the atmosphere that is responsible for the onset of extreme weather events.

Big Data/Machine Learning, Deep Learning, Fluid Modelling and Simulation

• No Major Restriction

Earth, Atmospheric, and Planetary Sciences

Lei Wang

Our research group is in the midst of constructing new quantum optics characterization setups. These setups are used to characterize the photoluminescence properties of different quantum emitters down to the single photon level! This single emitter level property characterization is critical in the development of quantum optical computing, sensing, and communications! We are looking for an undergraduate student that can help with the development of software to control the various parts of the setup and to write the drivers to interface the hardware to the upper-level analysis software required to control the setup.

Big Data/Machine Learning, Deep Learning, Fabrication and Robotics, Material Processing and Characterization, Nanotechnology, Other

• No Major Restriction

Experience with embedded systems programming, analog to digital conversion experience, driver experience, python software experience.

Electrical and Computer Engineering (ECE)

Vladimir Shalaev

This project is one of several SCALE SURF research projects, and is restricted to US citizens. If you are interested in more than one SCALE SURF project, you can apply to all of them with one application. ** Be sure to address each project by name in your application. ** See https://nanohub.org/groups/scale/research_su23 to view all of the SCALE SURF research projects for summer 2023.

Most of the materials we encounter in our daily lives are ‘bulk’ materials – they contain an enormous number of atoms in all three dimensions. However, if we instead consider materials with one dimension of only a few atoms in thickness, like graphene, we can achieve many unique physical and chemical properties unique from their bulk counterparts. For example, 2D magnetic materials have drawn significant attention because of their application in spintronics and quantum computing. One class of 2D materials with the potential to serve as the first room-temperature 2D magnets are MXenes, near atomically thin transition metal carbides or nitrides. For a magnetic material, the configuration can be ferromagnetic (FM) or antiferromagnetic (AFM) depending on the direction of spins of electrons. Using electronic structure calculations based on density functional theory (DFT), we can identify the magnetic configuration with lower energy. Further, the critical temperature, e.g. Curie temperature, is the temperature above which the material loses the spontaneous magnetization. For real-world applications, magnetic materials with a critical temperature that is higher than room temperature are desired. This project will combine DFT calculations to discover magnetic MXenes with high Curie temperatures.

In your application, please specify which of the SCALE technical areas you are most interested in. The technical areas are:
• Radiation Hardening
• System-on-Chip
• Heterogenous Integration/ Advanced Packaging
• Program Evaluation
Be sure to name any specific SCALE projects you are interested in, and include information about how you meet the required and desired experience and skills for each of these projects.

For US citizen students who are interested: you can become part of the Purdue microelectronics program called SCALE, sponsored by the Department of Defense. In SCALE, you will have opportunities for continuing research (paid or for credit) and industry and government internships throughout your time at Purdue. Please apply to SCALE here: https://research.purdue.edu/scale/.

Big Data/Machine Learning, Deep Learning, Material Modeling and Simulation, Microelectronics, Nanotechnology

• No Major Restriction

Introductory materials science or physics/chemistry of materials. Introductory programming

MSE

Alejandro Strachan

Two projects are available. One involves the investigation of enhancing optical imaging resolution using single photon counting techniques. Conventional optical imaging has a hard limit on its spatial resolution, to about one half of the wavelength, and many situations can benefit from higher resolution. In addition, it is challenging to image through scattering media. By way of example, being able to sense with light deeper in the brain would be of enormous benefit in neuroscience. The statistics of photons emitted by or transmitted through an object contain valuable information about the object which could be used to enhance image resolution and possibly see through substantial background scatter. Experiments will be conducted using laser light and with a set of single photon avalanche detectors (SPADs) to measure photon correlations in time, over wavevector (direction), and between detectors in various imaging configurations. Results from these experiments will be used to assess the effectiveness of various techniques for enhancing spatial resolution in imaging applications. This work has a diverse set of potential applications including biological imaging, sensing defects in semiconductors, and imaging through fog. The other project relates to experimental work and the modeling of optical forces on structured membranes induced by a laser. The mechanical motion of a thin membrane deflected by laser light will be used to determine the membrane properties from experimental and simulated data. This will allow extraction of the mechanical material properties and more generally the validation of a theory for optomechanics that can then be used in design. The nascent field of optomechanics offers enormous impact scope, including remote actuation and propulsion, of importance in fields as diverse and molecular biology, communication, and transport. This project relates to attaining the underpinnings to move along such paths in engineering, as well as the basic physics of optical forces in material at small length scales.

Big Data/Machine Learning, Biological Characterization and Imaging, Biological Simulation and Technology, Composite Materials and Alloys, Deep Learning, Material Processing and Characterization, Medical Science and Technology, Nanotechnology

• Electrical Engineering
• Mechanical Engineering
• Physics
• Biomedical Engineering

Students with an interest in experimental or modeling work and some background in electromagnetics would be a good fit for this project. The undergraduate student will work with graduate students to perform experiments in an optics laboratory, modeling, data analysis using MATLAB or python, and review relevant literature.

Electrical Engineering

Kevin Webb

Deep neural networks (DNNs) are widely used, from image recognition in autonomous vehicles to detecting anomalies in system logs. Training these networks incurs a huge carbon footprint. Reusing pre-trained neural networks (PTNNs) reduces this cost and improves engineering efficiency. However, little attention has yet been paid to improving the software engineering infrastructure that supports the trustworthiness of PTNNs. At present, PTNNs are shared across industry via model hubs: collections of PTNNs organized by problem domain and machine learning framework. These zoos imitate traditional software registries, such as NPM and Maven, whereby engineers share software packages. PTNNs are still in their infancy, and there are many unknowns regarding their trustworthy exchange between engineering teams.

Undergraduate student(s) will work with graduate students on projects related to analyzing PTNNs, developing tools to standardize them (e.g. ONNX), and developing tools to measure them.

Cybersecurity, Deep Learning

• No Major Restriction

Successful applicants should have most of the following: Introductory coursework or equivalent experience in machine learning and deep learning. Strong programming skills, familiarity with Linux programming environment (e.g. you are comfortable on the terminal). Vague knowledge of cybersecurity (e.g. buffer overflows). Knowledge of web systems (you know what React and Flask are, you've used one of them before). Data analysis skills (e.g. with Pandas). Successful applicants are likely EE, CompEng, or CS majors.

Electrical & Computer Engineering

James Davis