Summer Research Opportunities

Disruption of long-term memory by interictal epileptiform activity

• Background: IEDs have been shown to disrupt memory, but the mechanisms by which they do so are not clear. Hippocampal ripples order and replay neural firing, facilitating physiologic memory storage.
• Hypothesis: During learning, hippocampal ripples are decreased, or potentially replaced by IEDs, in epileptic animals, contributing to poor long-term memory.
• Methods: Analysis of rat hippocampal neurophysiologic data, detection of IEDs and ripples, analysis of rat behavior and memory during performance of a cheeseboard maze task (video).
• Goal: Decode neural signals that predict deterioration of memory.
• Contact: Dion Khodagholy ([email protected])

Responsive electrical stimulation to improve memory in epilepsy

• Background: Coupling of hippocampal IEDs to spindles in the medial prefrontal cortex (mPFC) is correlated with poor long-term memory.  Certain forms of electrical stimulation applied to the mPFC can abolish spindle oscillations.
• Hypothesis: Application of closed loop electrical stimulation in response to IED occurrence can prevent spindle oscillations and improve long-term memory.
• Methods: Analysis of rat in vivo neurophysiology and behavior for rats undergoing responsive electrical stimulation compared to control rats.
• Goal: Determine how responsive electrical stimulation affects neural networks and memory.  
• Contact: Dion Khodagholy ([email protected])

Characterizing neural networks in mouse models of pediatric epilepsy and evaluating response to targeted genetic interventions 

• Background: Although rodent models of pediatric epilepsies exist, analysis of in vivo neurophysiologic data during ages corresponding to the human neonate, infant, and child is lacking. This limits our ability to evaluate precision medicine therapies. 
• Hypothesis: Rodent models of pediatric epilepsies will exhibit altered oscillatory patterns and maturation compared to wildtype rodents that will permit longitudinal study and better understanding of epileptogenesis in these disorders. These parameters can be tracked to evaluate response to genetic therapy.
• Methods: Analysis of mouse developmental in vivo neurophysiology.
• Goal: Decode characteristic neural network features corresponding to emergence of epileptic activity across mouse models of pediatric epilepsy. Predict and track response to genetic therapies.
• Contact: Dion Khodagholy ([email protected])

Intelligent Spectrum Coordination for Future Wireless Networks

Students working on this project will construct SCMs based on real measurements of wireless signals, develop novel frequency coordination protocols based on SCMs, and implement the protocols in a custom-built python simulator and/or in a Software-Defined Radio testbed.

• Requirements: 
  • Some experience with Python and MATLAB is required. Some background in Wireless Communications is preferred. 
• Website
• Mentor: Gil Zussman
• Contact: Dr. Igor Kadota <[email protected]>

Weather Effects on High Frequency Communication Links

In this project, students are expected to learn how the mmWave radar works, design experiments to obtain labeled data, perform measurements, and develop the classifier.

• Requirements: 
  • Some background in Digital Signal Processing such as FFT is required.
    Some experience with Python is required.
  • Some experience with classification predictive modeling is preferred.
  • Some understanding of wireless networks, or interest to learn is preferred.
  • Must be able to access the Columbia Morningside Heights campus
• Website
• Mentor: Gil Zussman
• Contact: Dr. Igor Kadota <[email protected]>

Evaluating 5G mmWave Outdoors-to-Indoors

Students working on this project will be responsible for extensive data collection and analysis. The results can guide the design and deployment of next generation wireless networks.

• Requirements: Must be able to access the Columbia Morningside Heights campus. Some background in Wireless Communications is preferred.
• Website
• Mentor: Gill Zussman
• Contact: Abhishek Adhikari <[email protected]> 

Video analytics over edge-cloud networks 

Video analytics can be used in different applications, including traffic control, security surveillance, and factory floor monitoring. A typical video analytics application consists of a pipeline of video processing modules including an NN-based object detector. The pipeline has several knobs such as frame resolution, frame sampling rate, bitrate, and detector model. The choice of configuration impacts resource consumption, latency, bandwidth requirements, and accuracy of the video application. The best configuration for a video analytics pipeline also varies over time, often at a timescale of minutes or even seconds. In many cases, the policy that reduces the frame rate and lowers the resolution can save resources without impacting the accuracy. In this project, we design and evaluate a controller that dynamically picks the best configurations for existing NN-based video analytics pipelines.

• Requirements: 
  • Programming languages: C/C++, Python, Shell scripting
  • Computer vision: previous experience in working with  Deep learning models for object detection/tracking e.g. YOLOv4/Nvidia DCF tracker
  • Preferred: Familiarity with the GStreamer library, DeepStream SDK.  
• Website
• Mentor: Gill Zussman
• Position Dates: Spring May 31-Aug. 19
• Hours per Week: 35
• Position type: Hybrid (both remote and on site)
• Contact: Mahshid Ghasemi <[email protected]>

Projects in the Analog and RF Integrated Circuits Design Research Group

Analog vs Digital Feature Extraction for On-Device Keyword Spotting

The advent of ubiquitous computing calls for intelligent human-machine speech interfaces, in particular on-device keyword spotting (KWS)--think Amazon Alexa and Google Home. However, the ultimate goal is a KWS system that can survive on a coin cell battery for several years, which demands ultra-low power (nanowatt levels). Contemporary KWS systems consist of a frontend feature extractor and backend neural network, where the feature extractor is the power dissipation bottleneck. An open research question is whether feature extraction is more power-efficient using digital approaches fueled by Moore's Law or analog approaches inspired by history. Tackling this open-ended research question will entail traversing the entire stack of circuit design research skills: literature review, pencil-and-paper analysis, exploration in MATLAB, and analog & digital design in Cadence. Further, this work can potentially lead to co-authorship on a publication.

Phased Array Antenna Design for RF Sensing Systems

Emerging communication systems are relying on the use of multiple antenna systems and intelligent ways of allocating resources to increase channel capacity. Phased array antenna systems are being used in unique ways to solve new challenges in next-generation communication systems. In this project, you will work on the design, simulation, and testing of a phased array antenna that will be used to perform various sensing and communication tasks. The work will consist of literature review, MATLAB simulations, EM simulations, prototyping, PCB design, and antenna testing.

High Performance Electronics Under Cryogenic Conditions

The push towards quantum computing and deep-space exploration motivates the need for scalable electronics that can tolerate a wide temperature range, from near super-conducting (~4 K) up to 400 K.  The current focus is in understanding how CMOS circuits, particularly analog-to-digital converters done in a conventional 65nm process, respond to these temperature changes and what can be manipulated to compensate for any unwanted behavior.  However, before components can be tested at the target temperatures, a number of infrastructural tasks must be done to understand the experimental setup.  These tasks include but are not limited to (1) RTL coding, verification, and implementation for ADC data acquisition on an FPGA that can communicate data to a computer running a GUI.  (2) Setup and validation of a temperature-stable cryostat for liquid nitrogen and a hot-box, and (3) testing of COTS parts, such as RF balun transformers and passive components, at these temperatures. (4) Design of a temperature-stable switched-capacitor closed-loop amplifier for use in a pipelined SAR ADC.   

• Requirements: Please see Kinget Group website for materials to submit. 
• Website
• Mentor: Peter Kinget
• Contact: [email protected]

AI on the EDGE - Deep Learning with Low-Power GPU and IoT Devices 

Research with low power portable computing devices for data acquisition and AI-based processing. Applications in smart cities and healthcare. Hands-on experiments with NVIDIA Jetson Nano, Google Coral, Intel Neural Compute Stick. Sharing computational load with edge-cloud compute and data servers. Measurement of latencies and exploration of real time capabilities and constraints.

• Qualifications/skill-set: Signal Processing, Video Processing, Data Analysis, Software Development, some Machine Learning and Deep Learning.
• Hours per Week: 20-35 hours 
• Direct Supervisor: Zoran Kostic
• Paid Position: limited
• Credit: Can be up to 3 credit hours
• Number of positions: multiple

COSMOS Smart Intersections, Cloud-Connected Vehicles - Applications of Deep Learning

Applications of deep learning to project COSMOS.

COSMOS - advanced wireless research platform -> link, link, link, link.

Applications of Deep Learning - Research on: (i) COSMOS cloud connected vehicles, (ii) Monitoring of traffic intersections, using bird’s eye cameras, supported by ultra-low latency computational/communications hubs; (iii) Simultaneous video-based tracking of cars and pedestrians, and prediction of movement based on long-term observations of the intersection; (iv) Real-time computational processing, using deep learning, utilizing GPUs, in support of COSMOS applications; (v) Sub-100ms latency communication between all vehicles and the edge cloud computational/communication hub, to be used in support of autonomous vehicle navigation. The research is performed using the pilot node of the project COSMOS infrastructure https://cosmos-lab.org/.

COSMOS applications - short description: smart cities, smart intersections, cloud connected vehicles, applications of artificial intelligence, real time edge computing, https://cosmos-lab.org/experimentation/smart-city-intersections/.

• Direct Supervisor: Zoran Kostic
• Hours per Week: 20-35 hours Paid Position: limited
• Credit: Can be up to 3 credit hours
• Number of positions: multiple
• Qualifications/skill-set: Machine Learning, Deep Learning, Signal Processing, Video Processing, Data Analysis, Software Development, Cloud Computing, GPUs.

Using speech and language to identify patients at risk for hospitalizations and emergency department visits in homecare

This study is the first step in exploring an emerging and previously understudied data stream - verbal communication between healthcare providers and patients. In partnership between Columbia Engineering, School of Nursing, Amazon, and the largest home healthcare agency in the US, the study will investigate how to use audio-recorded routine communications between patients and nurses to help identify patients at risk of hospitalization or emergency department visits. The study will combine speech recognition, machine learning and natural language processing to achieve its goals. 

• Direct Supervisor: Zoran Kostic
• Hours per Week: 20-35 hours
• Paid Position: limited
• Credit: Can be up to 3 credit hours
• Number of positions: multiple
• Qualifications/skill-set: Signal Processing, Speech Processing, Data Analysis, Software Development, some background and interest in Cloud Computing, GPUs, Machine Learning, Deep Learning.

SensorHub - A Modular Platform for Rapidly-Deployable Sensing Systems

Researchers in various fields frequently find them spending too much time and budget in engineering a sensor system to collect the data they need. Tech-savvy homeowners dream to customize their smart home but are set back by its engineering complexity. The goal of this project is to create a Raspberry Pi based system with a modular interface to allow users from different engineering backgrounds to mix and match different sensors within seconds. We also plan to develop a customized indoor drone system, which, after integrating with this project, can load / unload sensors automatically and carry a specific sensor to a specific indoor location, with the control of a program.

• Requirements: 
  • Experience in Python programming, UNIX-based systems, PCB Design and embedded coding. 
  • Experience in designing mechanical systems and / or designing customized drones is preferred.
• Website
• Mentor: Prof. Xiaofan (Fred) Jianc
• Contact: Scott Zhao <[email protected]>

Data-driven At-home Mental Health Assessment and Intervention

The growth of IoT devices and sensors enables over 70% of homes in the United States to have at least one smart home device. Reports show that with the impacts of the pandemic, more than half of employed adults are currently working from home. The Census Bureau reports that there were 37 million one-person households in 2021. In addition, the pandemic makes the in-person therapist/doctor visit constraint and challenging. Thus leveraging the existing smart home devices to perceive the physical and mental status of the occupants would be a desirable solution to take care of the well-being of people who live alone or in the early phase of particular mental diseases. Besides, some smart home devices could also be utilized to provide appropriate interventions.

In collaboration with therapists, this project takes advantage of the prevalence of low-cost smart home sensors to bring mental health services and support to everyone’s home. This environment does not require patients to wear anything to avoid discomfort and won’t affect the patient’s regular life routine. The project aims to use smart home sensors devices such as the proximity sensor in Roomba and machine learning methods to predict mental health problems, to use devices such as Alexa to chat with the users to provide mental health support and to use actuators to provide intervention in case of an emergency. We want to provide objective evaluations to the patients for self-assessment or support materials for doctor visits.

• Requirements: 
  • Experience in Python programming and data analysis.
  • Experience in machine learning would be preferred.
• Website
• Mentor: Prof. Xiaofan (Fred) Jiang
• Contact: Jingping Nie <[email protected]>