Reza Abbasi-Asl, Professor

Closed (1) Stability driven interpretation and compression of neural networks

Applications for fall 2021 are now closed for this project.

Deep neural networks achieve state-of-the-art performance in many tasks such as computer vision and natural language processing. Interpreting deep networks is essential for applying them to scientific applications such as healthcare. Two prominent approaches to interpret neural networks are saliency methods and network compression. However, both methods could result in unstable or unreliable interpretations. In this project, we study the problem of stability in interpretation of neural networks. In particular, we use statistical techniques to build more interpretable neural networks and assess our methods in computer vision tasks.

The successful candidate will start by reading relevant papers, implement the basic interpretability frameworks for off-the shelf convolutional neural networks and quantify stability in these methods. Then use statistical principles to increase the stability in interpretations. Finally, the results will be documented and presented in form of a research publication. Note that the successful candidate will be working remotely on this project.

Qualifications: Candidate should have experience programming with Python and implementation of neural networks in platforms such as PyTorch, TensorFlow, Keras, or Caffe. Familiarity with convolutional neural networks and image processing, and medical image analysis is desirable but not essential.

Weekly Hours: 12 or more hours

Off-Campus Research Site: remote

Related website: http://abbasilab.org

Closed (2) Modeling and analysis of neural activity in visual cortex through the lens of machine learning

Applications for fall 2021 are now closed for this project.

Characterizing the the neural function in the brain and its relationship with connectivity is an eminent question of visual sensory processing. With the recent increase in the amount of the data collected from brain, tools based on machine learning principles play an essential role in understanding the brain function. The data collected from brain has a wide variety of modalities such as electrophysiology, calcium imaging, electron microscopy, and gene expression. In this project, we explore the functions in the brain via two distinct datasets: The first dataset is collected in collaboration with Allen Institute. It consists of visual responses from excitatory neurons within an 800X800 um region of primary visual cortex in mouse, spanning all visual layers from pia to white matter. This includes 750 2-photon and 35 3-photon calcium imaging planes spaced by ~16 um. The second dataset is electrophysiological recordings from visual cortex in macaque. This includes data from V1, V2, V4 and IT. Our goal is to examine the single-cell and population activity in visual cortex, build predictive models of spike rates and decoding models to reconstruct the visual experience from brain activity.

The successful candidates will start by reading relevant papers and learning about the dataset. Then the candidate will use our recently-developed analysis and modeling pipeline to study the dataset in more depth. This includes assessing multiple metrics, including receptive field profile, direction and orientation selectivity indices, reliability of response, signal and noise correlations, and sparseness of response. In another phase, the candidate will work on the predictive models based on convolutional neural networks to predict spike rates. Then an inverse model will be developed to reconstruct the visual experience from the neural activity. Finally, the results will be documented and presented in form of a research publication. Note that the successful candidate will be working remotely on this project.

Qualifications: Candidate should have experience programming with Python. Familiarity with computational neuroscience and convolutional neural networks is desirable but not essential.

Weekly Hours: 12 or more hours

Off-Campus Research Site: remote

Related website: http://abbasilab.org

Closed (3) Unsupervised pattern recognition in biomedical image data

Closed. This professor is continuing with Spring 2021 apprentices on this project; no new apprentices needed for Fall 2021.

Computational tools based on machine learning principles have shown promising results in analyzing medical images. Deep convolutional neural networks (CNNs) are one of the most successful tools in this domain. While most of the algorithms based on CNNs are supervised and with the increasing amount of un-labeled datasets, it is essential to design unsupervised or semi-supervised methods for analyzing medical images. In this project, we study unsupervised and semi-supervised methods to analyze two types of biomedical images from brain: 1. MR and fMRI images from humans and 2. Gene expression images in brain. We use tools such as deep autoencoders and non-negative matrix factorization (NMF) to find meaningful patterns in these images. The MRI dataset is from thousands of patients at UCSF as well as large-scale imaging datasets available online. The gene expression data is collected in collaboration with Allen Institute. This study could have valuable impact on relating patient health data to observed symptoms as well as help clinicians to easily benefit from the information hidden in the data.

The successful candidate will start by reading relevant papers and learning about the dataset. Then the candidate will implement tools based on deep autoencoders, NMF, and semi-supervised deep convolutional neural networks to analyze the data. The candidate will spend time on assessment of the patterns in the images identified through the algorithms and synthesize the results. Finally, the results will be documented and presented in form of a research publication. Note that the successful candidate will be working remotely on this project.

Qualifications: Candidate should have experience programming with Python and implementation of neural networks in platforms such as PyTorch, TensorFlow, Keras, or Caffe. Familiarity with convolutional neural networks, image processing, and medical image analysis is desirable but not essential.

Weekly Hours: 12 or more hours

Off-Campus Research Site: remote

Related website: http://abbasilab.org

Closed (4) Multi-modal large-scale patient data analysis via interpretable statistical and machine learning tools

Applications for fall 2021 are now closed for this project.

With the recent increase in the amount of the healthcare data, computational tools play an important role in bridging patient data to the disease symptomes. Statistical and machine learning principles are some of the tools that successfully have been used in this domain. However, interpretability of these tools has not been widely studied. in this project, we aim to build interpretable tools to extract information from large-scale multi-modal patient data. The data consists of structured clinical documents including lab measurements, metrics from imaging modalities, biographical information, etc. Both supervised and unsupervised tools are of interest. This project could lead to significant increase in our knowledge about neurological disease and how the patient data could predict the disease symptoms.

The successful candidate will start by reading relevant papers and learning about the dataset. Then the candidate will implement tools based on deep autoencoders, NMF, and semi-supervised deep convolutional neural networks to analyze the data. These tools will be used to analyze multi-modal data from patients with neurological disease. The candidate will spend time on assessment of the patterns in the data identified through the algorithms and synthesize the results. Finally, the results will be documented and presented in form of a research publication. Note that the successful candidate will be working remotely on this project.

Qualifications: Candidate should have experience programming with Python and implementation of statistical methods such as regularized regression and neural networks in platforms such as PyTorch, TensorFlow, Keras, or Caffe. Familiarity with convolutional neural networks, image processing, and medical image analysis is desirable but not essential.

Weekly Hours: 12 or more hours

Off-Campus Research Site: remote

Related website: http://abbasilab.org