Center for Undergraduate Research in Viterbi Engineering (CURVE) Fellowship

Project Title: Continuous Scene Representations for Agent Reasoning

Faculty / PI: Jesse Thomason

Lab Name: Grounding Language in Actions, Multimodal Observations, and Robots

Project Title: Continuous Scene Representations for Agent Reasoning

Faculty / PI: Jesse Thomason

Lab Name: Grounding Language in Actions, Multimodal Observations, and Robots

Department: Computer Science

Research Website: http://glamor.rocks/

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: This project will involve research activities to broaden Continuous Scene Representations (CSR) [https://prior.allenai.org/projects/csr] to facilitate natural language instructions. While existing CSR represents only visual information about the world, a linguistically-informed CSR would also encode information from language instructions. For example, seeing that a mug is on top of a table in a particular configuration is a specific, visual observation, but the instruction "Pour coffee into the mug on the table" implies there is some mug on some table in some configuration, a broader set of possibilities than the visual observation. By considering such linguistic implications and object affordances (e.g., mug can hold liquid, table can support solid objects), we envision applying CSRs to control virtual agents and to facilitate sim-2-real transfer to real robot systems.

Student Responsibilities: Working alongside a PhD student to generalize CSRs, likely initially in a simulation environment like AI2-THOR or iGibson, and possibly adapting on a physical robot platform such as the LoCoBot. Programming will be mostly or entirely in Python, and familiarity with deep learning libraries like PyTorch will go a long way. Having taken an AI or ML class before will help get started quicker.

Interview Required? Yes

Skills and Competencies (Preferred)

• Familiarity with Python programming
• Working knowledge of AI/ML concepts (e.g., have taken a class)
• Willing to dive deep when debugging (e.g., sometimes you gotta dig into the codebase of the library you're using to see what's going wrong)

Project Title: Continual Learning for Vision-and-Language Tasks

Department: Computer Science

Research Website: http://glamor.rocks/

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: This project will involve building new ML models and algorithms on top of the recently-released CLiMB benchmark (https://arxiv.org/abs/2206.09059), a benchmark to develop multimodal models that can learn continually. Multimodal models are a class of Machine Learning models that can take both images and text as input (i.e. they can see and they can read), and make predictions based on these inputs. For example, in the Visual Question Answering task, models are given an image and a question about the image, and the model must produce an answer by understanding both the image and question together. CLiMB applies these multimodal models to the challenging task of continual learning - where a model must learn to perform tasks one after another while not forgetting tasks that it has already learned. Continual learning is an important but relatively unexplored challenge in machine learning, particularly in the multimodal domain, which means that developments on this benchmark would be at the cutting edge of Machine Learning research.

Student Responsibilities: Working alongside a PhD student to add new models to the CLiMB benchmark, and develop new continual learning algorithms that can facilitate knowledge transfer across tasks while mitigating forgetting issues. Programming will be entirely in Python, so it is essential that the student must be comfortable programming in the same. The student must also have some working knowledge of Machine Learning (even if they are unfamiliar with state-of-the-art methods), and should have some experience with implementing ML models. Familiarity with the PyTorch deep learning library, and prior experience in NLP, is recommended but not necessary.

Interview Required? Yes

Skills and Competencies (Preferred)

• Familiarity with Python programming
• Working knowledge of AI/ML concepts (e.g., have taken a class)
• Willing to learn about NLP and multimodal learning (if unfamiliar), and read papers about state-of-the-art models and algorithms
• Willing to dive deep when debugging (e.g., sometimes you gotta dig into the codebase of the library you're using to see what's going wrong)

Project Title: Dense Haptics

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website:

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: DenseHaptics is a research platform for hapticians to explore novel interactions, to rapidly prototype and test haptic experiences. This platform creates a easy to use soundcard-based interface to deliver pure analog output that is amplified to support up to 40 actuators with a power output of up to 3W. This platform is portable and can support a variety of vibrotactile actuators. It can be used with multiple sets of such platforms to support an even larger array of actuators in a scalable cost effective manner.

Student Responsibilities: Software and hardware prototyping

Interview Required? Yes

Skills and Competencies (Preferred)

• C++
• Laser Cutting
• Python
• 3D printing
• Fabrication

Project Title: Design of Customizable Haptic Wearable System

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website: https://sites.usc.edu/culbertson/

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: In experimental haptics research, a typical part of the research pipeline is to create a harness to attach actuators to flexibly to different parts of the body. The project aims to create a platform of objects that can be rapidly modified to create custom harnesses via 3D printed objects, magnets and tubing to ensure that we can create a easy to prototype, easy to clean, flexible sized harnesses for testing haptic prototypes for actuation. The system also focuses on rapid donning and doffing.

Student Responsibilities: -To prototype and test various harness systems

-Use 3d printing and laser cutting to fabricate harness

-Design and iterate over several designs

Interview Required? Yes

Skills and Competencies (Preferred)

• CAD
• 3D printing

Project Title: Electrical Stochastic Resonance

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website:

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: Stochastic resonance (SR) is a phenomenon in which a signal that is normally too weak to be detected by a sensor, can be boosted by adding white noise to the signal, which contains a wide spectrum of frequencies. This project aims to replicate the phenomenon via Electrically induced Stochastic Noise to the hand such that the individual is potentially better at discerning virtual textures.

Student Responsibilities: Electronic prototyping, software development in C++ / python and hardware systems.

Interview Required? Yes

Skills and Competencies (Preferred)

• C++
• Electronics Prototype
• 3D printing
• Laser Cutting

Project Title: Rapid Modelling of Haptic Buttons

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website:

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: Haptic modelling of real-world objects like switches, sliders, rotary dials are challenging. These are modelled using methods like Finite element analysis or via complex physics simulations. This project aims to create an approximate model of the physical interaction by leveraging 3D printed magnet-based devices like Nintendo Labo, Mechamagnets to create an environment to capture the rapid changes in magnetic flux. This change in magnetic flux can be translated into a model to replicate the real-world objects in virtual environments.

Student Responsibilities: Data analysis, programming and fabricating electromechanical devices

Interview Required? Yes

Skills and Competencies (Preferred)

• Electronic Prototyping
• Matlab
• Python
• VR programming (Unity 3D)

Project Title: Tangible Block-Based Coding for Visually Impaired Individuals

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website: https://sites.usc.edu/culbertson/

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: The project is to create a system to allow visually impaired individuals to code using tangible blocks. These blocks will be 3D printed with a raised pattern on top indicating the coding concept (loop, variable, etc.). The blocks must be able to be recognized by touch alone, and will be placed on a board to write code using a custom block-based programming language. The system will be tested in a human subject study to evaluate its ease of use and effectiveness at teaching coding concepts. Relevant field experience: computer science, electrical engineering, mechanical engineering.

Student Responsibilities: The student will be responsible for completing a prototype and conducting a human subject study to evaluate the usability of the system. They will attend weekly lab meetings, regular weekly meetings with the PI and with the PhD mentor, and possibly journal club. The student will also be proactive with communicating when they are stuck in their work or need help, what their goals and expectations are, and for keeping up with literature (background readings will be provided to start).

Interview Required? Yes

Skills and Competencies (Preferred)

• Progamming
• Electronics
• 3D printing

Project Title: Variable Spring Haptics

Faculty / PI: Heather Culbertson

Lab Name: HaRVI Lab

Department: Computer Science

Research Website:

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: In Haptics, and specifically in Virtual Reality (VR) related haptics, deform-able objects post a challenge. This challenge can be a physical (hardware challenge), software (rendering) challenge to produce real-time deform-able objects. Most of the attempts use a motor that exerts a certain amount of force to counter act the grasping or compressive loads provided by the fingers on the haptic device. Some novel implementations uses breaking mechanisms to provide rigidness and softness to these virtual objects.  This project aims to create a system in which we can dynamically reduce the length of the spring thereby controlling the effective spring constant to create dynamic objects in VR.

Student Responsibilities: Prototype and help in developing the system

Interview Required? Yes

Skills and Competencies (Preferred)

• Laser Cutting
• Mechanical Engineering
• Programming (C++ and Python)
• 3D designing and 3D printing
• Electronic Prototyping

Project Title: Dominance Checking in Multi-Objective Search

Faculty / PI: Sven Koenig

Lab Name: Intelligent Decision Making Group

Department: Computer Science

Research Website: idm-lab.org

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: Multi-objective search is about finding a path in a graph whose edges are labeled with two costs instead of one. A graph could, for example, represent a road network, and its edges are labeled with the travel time and gas consumption needed to traverse them. The process of dominance checking (= checking whether one path is better than another) plays a key role for the efficiency of multi-objective search algorithms but can be implemented with different algorithms and data structures. The student will therefore study the efficiency of multi-objective search algorithms for different implementations of dominance checking.

Student Responsibilities: The student will read relevant literature, implement different ways of dominance checking, design and perform experiments, and write up the results. We aim for a conference publication of the results.

Interview Required? Yes

Skills and Competencies (Preferred)

• Good understanding of algorithms and data structures
• Good coding skills
• Interest in artificial intelligence and theoretical computer science

Project Title: Incremental Search  for Multi-Agent Path Finding

Faculty / PI: Sven Koenig

Lab Name: Intelligent Decision Making Laboratory

Department: Computer Science

Research Website: idm-lab.org

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: Multi-Agent Path Finding (MAPF) is the problem of finding collision-free paths for a set of agents on a graph, where each agent has unique start and goal locations and time is discretized into timesteps. MAPF has many applications such as for automated warehouses and traffic management. Conflict-Based Search (CBS) is one of the state-of-the-art complete and optimal algorithms that solve MAPF. CBS finds a path for each agent individually first and resolves collisions afterwards. To resolve a collision between a pair of agents, CBS adds a constraint to one of the agents and replans its path in order to satisfy the additional constraint. However, current implementations of CBS use spatial-temporal A* to replan the agent from scratch, which lowers its efficiency. In this project, we will replace A* with other incremental search algorithms (e.g., X* and D*) such that CBS only needs to replan a path segment instead of the whole path to satisfy the additional constraint.

Student Responsibilities: The student will read the relevant literature, implement an approach, design and perform experiments and write up the results. We aim for a conference publication of the results.

Interview Required? Yes

Skills and Competencies (Preferred)

• proficiency in C++ and Python (knowledge of ML packages is NOT required)
• basic knowledge of artificial intelligence (comparable to CS360 - Introduction to AI)
• will take at most 3 courses in Fall 2022 and can spent at least 2 hours in the lab per week

Project Title: Multi-Agent Reinforcement Learning

Faculty / PI: Sven Koenig

Lab Name: IDM Lab

Department: Computer Science

Research Website: http://idm-lab.org/project-p.html

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: Multi-agent path finding is about finding collision-free paths for a team of agents that move on a grid, see mapf.info. Applications include warehouse robots and game characters. In this project, the student will work on multi-agent reinforcement learning (MARL) for the multi-agent path finding (MAPF) problem. The project will focus on using MARL to generate solutions to small MAPF instances (e.g., finding collision-free paths for 10 agents on a 5x5 or 10x10 empty grid). The motivation for this project is to explore if a reinforcement-learning policy can serve as a high-quality lookup table to provide solutions to small MAPF instances and, therefore, be used as part of search algorithms that solve larger MAPF instances.

Student Responsibilities: The student will survey state-of-the-art MARL methods and implement reinforcement-learning algorithms. They will also evaluate the learned policies and integrate them into MAPF solvers. We aim for a conference publication of the results.

Interview Required? Yes

Skills and Competencies (Preferred)

• Minimal: (1) be proficient in Python and C++. (2) Have taken basic courses in AI, ML, and algorithms.
• Preferred: Junior or senior students with research experience in AI/ML/RL.
• It is important for the student to be able to devote sufficient time for the project.

Project Title: Designing and Evaluating Idle and Re-Engagement Behaviors in a Robotic Study Companion for Students with ADHD

Faculty / PI: Maja Mataric

Lab Name: Interaction Lab

Department: Computer Science

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: Prior studies have shown that college students with ADHD respond positively to simple social robots that monitor attention and provide immediate feedback during periods of inattention or impulsivity. Designing appropriate, non-invasive re-engagement behaviors is vital for creating a non-distracting study environment. This work aims to develop natural re-engagement behaviors in a simple, low-cost social robot. We will evaluate these behaviors by measuring their effect on students' engagement as they complete a schoolwork task.

Student Responsibilities: Students will use python to program idle and re-engagement behavior sequences for a Blossom socially assistive robot. They will design a user study to evaluate these behaviors during a schoolwork task.

Interview Required? No

Skills and Competencies (Preferred)

• Python Programming
• Basic mechanical + electrical skills
• Data processing

Project Title: Modeling Fine-Grained Cognitive-Affective Behaviors Of Children With Autism Spectrum Disorder (ASD) From Egocentric Robot Vision

Faculty / PI: Maja Mataric

Lab Name: Interaction Lab

Department: Computer Science

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: Current computer vision research doesn’t sufficiently focus on the social or autonomous behaviors necessary for successful human-robot interaction for children with ASD. Our work aims to introduce a large-scale dataset for modeling cognitive-affective states of children with ASD from an egocentric robot vision. We analyze the effects of multiple spatial processing models, video frame sampling strategies, and training input strategies in order to develop the best accuracy output for our dataset.

Student Responsibilities: Students will work analyzing the effects of multiple spatial processing models, video frame sampling strategies, and training input strategies in order to develop the best accuracy output for our dataset.

Interview Required? Yes

Skills and Competencies (Preferred)

• Python
• Deep Learning
• Machine Learning

Project Title: Towards Automatic Generation of Personalized Backchanneling Behaviors in A Socially Assistive Robot

Faculty / PI: Maja Mataric

Lab Name: Interaction Lab

Department: Computer Science

Research Website:

# of Freshmen Positions: 4

# of Continuing Student Positions: 6

Project Description: Backchannels (utterances such as “uh-uh” or nodding during a conversation) are critical to active listening and demonstrated engagement. Prior research has demonstrated that backchannel frequency follows different rules for various languages, and that people tend to follow the backchanneling patterns of their dominant language even when speaking in another language. This project aims to use machine learning to detect patterns for backchanneling frequency to create personalized models for individuals who come from different language backgrounds. These models will then be tested on a Socially Assistive Robot (SAR) in a future user study.

Student Responsibilities: Students will assist with design and implementation of user studies and annotation of collected participant data. They will also have the opportunity to learn how to use machine learning tools to develop affect models and gain familiarity with fundamental robotics software tools.

Interview Required? No

Skills and Competencies (Preferred)

• Python coding experience
• Curiosity about the academic research process
• Professionalism demonstrated by a strong sense of personal motivation, communication skills, and time management.
• interest and/or experience with machine learning

Project Title: Your Voice of Mindfulness: Evaluating and Personalizing Text-to-Speech Voices for Mindfulness Practice

Faculty / PI: Maja Mataric

Lab Name: Interaction Lab

Department: Computer Science

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: Past literature has shown that mindfulness-based exercises can be effective in lowering stress, anxiety and depression levels. However, physical and financial barriers prevent the provision of in-person mindfulness exercises for users from all socio-economic backgrounds. To lower these barriers, past work has studied alternative methods to administer mindfulness exercises, such as mobile applications (apps), conversational agents, and socially assistive robots. To ensure that these computer-based methods can deliver mindfulness exercise content with appropriate levels of quality, several major research questions remain unaddressed.

In this work, our objective is to study the effect of voice in mindfulness exercises. We aim to evaluate both human and computer-synthesized text-to-speech (TTS) voices in mindfulness exercises with and without a social embodiment of the voice, like a voice assistant or an assistive robot.

Student Responsibilities: Students will help design studies and perform statistical analysis to understand: 1) which features of the voice are preferable and suitable for administering mindfulness exercises; 2) how we can design better personalized computer-synthesized voices for improving computer-administered mindfulness exercises.

Interview Required? Yes

Skills and Competencies (Preferred)

• Python
• Statistical Analysis
• Study Design

Project Title: Robust and Miminally Supervised Knowledge Extraction from Natural Language Tech

Faculty / PI: Muhao Chen

Lab Name: Language Understanding and Knowledge Acquisition Lab

Department: Computer Science

Research Website: https://luka-group.github.io/research.html

# of Freshmen Positions: 0

# of Continuing Student Positions: 3

Project Description:

1. Knowledge acquisition tasks (e.g., relation extraction, entity and event extraction and typing, consolidation) face challenges including extreme label spaces, insufficient annotations and out-of-distribution prediction. To this end, we study the method for leveraging indirect supervision signals from auxiliary tasks (e.g., natural language inference, abstractive summarization, etc.) to foster robust and generalizable inference for (open-domain) knowledge acquisition or information extraction. In the same context, study the method for generating semantically rich label representations based on either gloss knowledge or structural knowledge from a well-populated lexical knowledge base, in order to better support learning with limited labels.
2. Knowledge graphs (KGs) provide both open-world and domain-specific knowledge representations that are integral to many AI systems. However, constructing KGs is usually very costly and requires extensive effort. A widely attempted solution is to learn knowledge acquisition models that automatically induce structured knowledge from unstructured text. However, such models developed through data-driven machine learning are usually fragile to noise in learning resources, and may fall short of providing reliable inference on large, heterogeneous real-world data. We are developing a general meta-learning framework that seeks to systematically improve the robustness of learning and inference for data-driven knowledge acquisition models. We seek to solve several key problems to accomplish the goal: (i) How to identify incorrect training labels and prevents overfitting on noisy labels; (ii) how do detect invalid input instances in inference (e.g., out-of-distribution ones) and allow prediction with abstention; (iii) automated constraint learning that ensures inference with global consistency; (iv) model robustness against noise perturbation; (v) mitigating spurious correlation of models captured in biased training.
3. Natural language understanding and generation tasks need to handle equivariance properties in data. For example, the narrative structure of an article can be reorganized, while still presenting the same content. In constrained NLG tasks with structural priors (e.g. data-to-text generation tasks), the structure of the prior can also be modified while presenting semantically equivalent content. However, existing sequential modeling of Transformer LMs cause downstream information extraction and NLG systems to be brittle to content-neutral transformations of input data. We are developing equivariance learning methods that allow Transformer models to give consistent output under any such content-neutral perturbations on input data.

Student Responsibilities: Lead or participate in research projects that lead to publications in top NLP conferences

Interview Required? No

Skills and Competencies (Preferred)

• Huggingface
• PyTorch

Project Title: STPLS/OWT

Faculty / PI: Andrew Gordon

Lab Name: Modeling & Simulation

Department: USC ICT

Research Website: https://ict.usc.edu/research/labs-groups/modeling-simulation/

# of Freshmen Positions: 5

# of Continuing Student Positions: 3

Project Description: Efforts from both academia and industry have adopted photogrammetric techniques to generate visually compelling 3D models for the creation of virtual environments and simulations. However, such generated meshes do not contain semantic information for distinguishing between objects. To allow both user- and system-level interaction with the meshes, and enhance the visual acuity of the scene, classifying the generated point cloud and the associated mesh is a necessary step. The objective of this research is to lay the groundwork for the semantic labeling of 3D point and clouds/meshes in integrating the proposed framework into the existing workflow for the creation of virtual environments and simulations. The proposed project is part of the One World Terrain (OWT) Project. One of the objectives of the OWT is to provide small units with the organic capability to create geo-specific virtual environments for training and rehearsal purposes to support military operations.

Student Responsibilities:

1. learn basic knowledge about ML/DL
2. test different ML/DL algorithms for the classification/segmentation tasks
3. create ground truth data for performance analysis

Interview Required? No

Skills and Competencies (Preferred)

• comfortable navigating in 3D environment
• knowledge with Python is preferred but not required
• knowledge with ML/DL is preferred but not required

Project Title: Outdoor Augmented Reality Tracking

Faculty / PI: Ramesh Govindan

Lab Name: Networked System Lab

Department: Computer Science

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: Augmented reality is an immersive experience in which the real-world environment is enhanced by computer generated information. One key component of augmented reality is the accurate registration of the real and the virtual object. The accurate registration heavily requires the capability of accurate localization and tracking. Today, the consumer level AR/VR headset relies on visual and inertial information to provide the tracking and localization. The current techniques work reasonably well in the indoor environment, but not in the outdoor environment, where the situation is more dynamic and visual information is less reliable. The more advanced visual tracking algorithms are being developed by the computer vision community, but the computing resources that are required by these algorithms are unaffordable on the AR headset.

This project is approaching this problem by integrating LiDAR sensors and edge computing. LiDAR can provide an accurate 3D position of objects and help localize the AR headset. Edge computing enables us to offload high demanding computation off from the headset, but still maintain a short latency.

Student Responsibilities: Develop AR/VR applications in Unity/Unreal/native environment that render the situational information of the environment.

Learn how to operate sensors (LiDARs, GPS, cameras, etc) and develop programs to collect data.

Be a part of the team and communicate well with the mentors.

Interview Required? Yes

Skills and Competencies (Preferred)

• AR/VR development experience
• Knowledge in C++
• Computer Vision

Project Title: Volumetric Video

Faculty / PI: Ramesh Govindan

Lab Name: Networked Systems Lab

Department: Computer Science

Research Website: https://nsl.usc.edu/

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: Volumetric video is a next-generation 3D video for augmented and virtual reality (AR/VR) that allows the viewer to change the viewpoint in 6 degrees of freedom. We are looking for undergraduate students who will work in NSL (Networked Systems Laboratory) to build research prototypes for the following projects:

- End-to-end volumetric video streaming for applications like video conferencing and virtual classrooms.

- 3D video analytics using machine learning on volumetric video.

Student Responsibilities: Students will be exposed to cutting-edge tools for research and development in the AR/VR space. They will be responsible for the following:

- Develop faster algorithms that reduce latency for 3D capture and rendering.

- Use machine learning models that can work in real-time.

- Work with Ph.D. students on an existing codebase to integrate the above algorithms.

- Read existing literature in related areas.

Interview Required? Yes

Skills and Competencies (Preferred)

• Programming in C++ and Python
• XR development with Unity and Oculus
• Machine learning tools like Pytorch or Tensorflow
• Interested in learning about 3D data like point clouds and meshes

Project Title: Cyber-physical fuzzing simulation framework

Faculty / PI: Luis Garcia

Lab Name: STEEL

Department: Computer Science/Information Sciences Institute

Research Website: https://lagarcia.us

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: Fuzzing is a standard automated technique for testing software for unwanted or unexpected behavior against different software inputs. When we put software on cyber-physical systems, i.e., systems that interact with the physical world through sensors and actuators, fuzzing the software is not sufficient for triggering unwanted behaviors. The sense-to-actuate pipeline is vulnerable to cyber-physical side-channels, such as sensor spoofing or false data injection. Conversely, we cannot just subject the physical systems to various signal generators as the systems would be subject to damage and could be costly. If we opt for simulation, modeling these behaviors would require capturing the sensor and actuator physical dynamics with high fidelity. The models would also need to understand how the sensor and actuator signals interact with the software. Students will study existing techniques to capture sensor and actuator dynamics in simulation and design and implement a prototype on an existing cyber-physical system testbed that can incorporate these models.

Interview Required? Yes

Skills and Competencies (Preferred)

• Ability to work dynamically with various programming languages
• Python
• Basic control systems/signal processing
• AI for sensor data
• Good software engineering/development practices

Project Title: Decoding how humans encode memory

Faculty / PI: Luis Garcia

Lab Name: STEEL

Department: Computer Science/Information Sciences Institute

Research Website: https://lagarcia.us

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: Proj. 1: Recent advancements in closed-loop deep brain stimulation (DBS) have enabled more intelligent autonomy for therapeutic intervention across a wide range of neurologic and psychiatric disorders. The predominant approach relies on control-theoretic approximations of the brain’s complex functional relationships with the external environment–-in particular, a mapping between targeted stimulation and naturalistic responses of different regions of the brain. However, existing approaches fail to capture the environmental context of neuronal biomarkers. Students will explore mechanisms to reason about complex sensor data from our recording platform, NeurIoT, that leverages a comprehensive set of Internet-of-Things (IoT) sensors to capture the human experience and environmental context, i.e., a subset of human sensory channels, in order to estimate the state of the human brain and provide the foundation for smarter, context-dependent DBS.

Interview Required? Yes

Skills and Competencies (Preferred)

• Ability to work dynamically with various programming languages
• Python
• Basic control systems/signal processing
• AI for sensor data
• Good software engineering/development practices

Project Title: Extracting Math Expressions from Low-level Binaries for Reverse Engineering

Faculty / PI: Luis Garcia

Lab Name: STEEL

Department: Computer Science/Information Sciences Institute

Research Website: https://lagarcia.us

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: 

• Build upon prior reverse engineering tools to extract math expressions
• Use natural language models to abstract low-level expressions
• Synthesize lots of CPS binaries and tie to them to dynamic simulators

Interview Required? Yes

Applied skills: CPS modeling, binary analysis experience (static and dynamic), C/C++, simulation development

Project Title: Neural-symbolic Encrypted Network Traffic Classification

Faculty / PI: Luis Garcia

Lab Name: STEEL

Department: Computer Science/Information Sciences Institute

Research Website: https://lagarcia.us

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: 

• Knowledge graphs to taxonomize network traffic characteristics
• Stich together neural and symbolic interfaces to optimize network traffic classification

Interview Required? Yes

Applied skills: general deep learning/ML, neurosymbolic programming, knowledge graphs

Project Title: Simple, Explainable, and Efficient Malware Detection

Faculty / PI: Luis Garcia

Lab Name: STEEL

Department: Computer Science/Information Sciences Institute

Research Website: https://lagarcia.us

# of Freshmen Positions: 1

# of Continuing Student Positions: 1

Project Description: 

• Natural language processing to sift through malware reports and taxonomize malware behavior via knowledge graphs
• Construct scalable malware analysis techniques to identify malware signatures

Interview Required? Yes

Applied skills: knowledge graphs, NLP, malware analysis

Project Title: AI for Biodiversity Conservation

Faculty / PI: Bistra Dilkina

Lab Name: USC Center for AI in Society

Department: Computer Science

Research Website: https://www.cais.usc.edu/projects/conservation-sustainability

# of Freshmen Positions: 0

# of Continuing Student Positions: 4

Project Description: We have several biodiversity conservation projects that students can join:

- ML and network interdiction optimization models to stop wildlife trafficking

- data mining and visualization of large number of biodiversity and environmental projections under climate change across large spatial areas (megaSDM)

- helping conersvationists decide how to place/layout cameras across protected areas to monitor vulnerable species

Student Responsibilities: - collect, analyze and visualize geospatial data using Python (gis data like rasters and network files)

- collect/collate and visualize data for the projects (GIS, Python, networks)

- build ML models or optimization algorithms to support the goals of the project

- present research progress at weekly lab meeting

- write a report / paper describing the project and results

Interview Required? yes

Skills and Competencies (Preferred):

• Python (needed)
• Good communication skills (needed)
• Algorithms course (preferred)
• GIS/geo data experience (preferred but not required)
• ML (preferred)
• visualization

Project Title: Machine Learning for Treatment of Substance Use

Faculty / PI: Bistra Dilkina

Lab Name: USC Center for AI in Society

Department: Computer Science

Research Website: https://www.cais.usc.edu/projects

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: We are working with collaborators from the School of Social Work and the Keck School of Medicine to develop machine learning methods that help them study the underlying risk and protective factors in substance use (opioids, stimulants, alcohol, etc). In addition to studying most important features, these models can inform health providers when a patient is likely to relapse to using harmful substances and help them intervene successfully.

Student Responsibilities: Develop classification and regression ML models based on health data, socio economic data from Census, and environment / neighborhood data to predict relevant health outcomes and study feature importance. Work closely with interdisciplinary team and give presentations on results.

Interview Required? Yes

Skills and Competencies (Preferred)

• Python
• Machine Learning (Classification / Regression)
• Good Presentation Skills

Project Title: Predict Child Malnutrition in Kenya

Faculty / PI: Bistra Dilkina

Lab Name: USC Center for AI in Society

Department: Computer Science

Research Website: https://www.cais.usc.edu

# of Freshmen Positions: 0

# of Continuing Student Positions: 2

Project Description: This project is a collaboration between CAIS (the USC Center for AI in Society) and IIGH (the USC Institute for Inequalities in Global Health). This project aims to help address malnutrition in children under five throughout Kenya by developing a spatio-temporal machine learning model to more accurately predict anticipated risk of famine and malnutrition among children. This AI model will leverage data from the national health information system, along with remote sensing, weather and agricultural data, to forecast when and where food scarcity will result in increased cases of malnutrition among children. Data from this AI model will inform a food and nutrition dashboard intended for use by health facilities and civil society organizations that provide nutrition interventions throughout the country. Our goal is to provide a data-based solution to improve the efficiency, accuracy and preparation of nutrition supplementation and treatment services to ensure the most vulnerable children in Kenya receive the nutrition needed to thrive.

Student Responsibilities: The student will be the main ML model developer for this project, and will need to process the health statistics data, spatial data on health facility locations and as well as other data sources. The student will then develop spatio-temporal machine learning regression models to predict number of malnutrition cases each month at each facility. The student will need to work with interdisciplinary team and give presentations on results. Keen interest in advanced machine learning methods is highly encouraged.

Interview Required? Yes

Skills and Competencies (Preferred)

• Python - needed
• Strong communication/Presentation skills - needed
• Machine Learning (Regression, tabular data) – needed
• Deep Learning, GCNs, Graphical Models – preferred
• Microsoft Azure experience – strongly preferred

Additional Details:

Work with Microsoft AI for Good research lab, the Keck School of Medicine, Kenyan NGOs and Government partners.

Project Title: Using AI for Disaster Mitigation Planning

Faculty / PI: Bistra Dilkina

Lab Name: USC Center for AI in Society

Department: Computer Science

Research Website: https://cais.usc.edu/projects/ai-for-disaster-planning-and-response

# of Freshmen Positions: 0

# of Continuing Student Positions: 1

Project Description: The project develops methods to help agencies assess the impacts of disasters on society and decide where to strategically upgrade critical infrastructure such as the urban water pipe network or the road network in order to improve our resilience to disasters like earthquakes and floods.

Student Responsibilities: The student will write code and analyze impact of earthquakes and floods on the infrastructures networks across different cities, will use optimization approaches like Integer Programming, and machine learning approaches to identify relationships between socio-economic factors and disaster exposure. The ideal student will. have good Python skills. Background either in optimization (ISE) and/or machine learning and/or GIS skills (dealing with large shapefiles in Python), or eagerness to learn these skills.

Interview Required? Yes

Skills and Competencies (Preferred)

• Python - needed
• GIS, spaital analysis - preferred
• Running scripts on the High-Performance Comuting cluster, some parallelization
• Algorithms