Past events

Graduate Programs Online Information Session

RSVP today!.

During each session, the graduate staff will review:

Requirements (general)
Applying
Prerequisite requirements
What makes a strong applicant
Funding
Resources
Common questions
Questions from attendees

Students considering the following programs should attend:

University closed

The University of Minnesota will be closed in observance of Memorial Day.

View the full schedule of University holidays.

Application deadline for data science major

The application deadline for data science majors is May 25.

Undergraduate students typically apply to a major while enrolled in fall semester courses during their sophomore year (third semester).

Submit your application here.

All applicants will be notified of their admission decision via email within three weeks of the application deadline.

Department of Computer Science & Engineering Undergraduate Student Graduation Event

In addition to celebrating you at the 2022 commencements in CLA and CSE, our department will be hosting a department-specific graduation event to celebrate our graduating students, and you are invited! This event will be held on Thursday, May 12th, from 9:00 am - 11:00 am (remarks from the department will begin at 10:00) at the University of Minnesota Recreation and Wellness Center in the Beacon Room on the second floor.

Please RSVP for this event if you graduated recently, or you will graduate soon. Even if you are not able or planning to attend the event, you are still welcome to submit the RSVP form as a graduating student.

Caps and gowns are optional for our departmental event. However, we recommend dressing professionally if you plan on attending without a cap and gown. Masking is strongly recommended for the event but is no longer required. Light food and refreshments will be provided.

End of spring semester

The last day of the spring 2022 semester is Wednesday, May 11.

View the full academic schedule on One Stop.

Final exams begin

Final exams for spring 2022 will be held between Thursday, May 5 and Wednesday, May 11.

Monday, May 3 and Tuesday, May 4 are study days.

View the full academic schedule on One Stop.

Last day of instruction

The last day of instruction for the spring 2022 semester is Monday, May 2.

View the full academic schedule on One Stop.

Department of Computer Science & Engineering Graduate Student Graduation Event

RSVP Link

Our Graduate Student Graduation Celebration event will be held Friday, April 29th from 9:00 - 11:00 (remarks from the department at 10:00 am) held in the University of Minnesota Recreation and Wellness Center Beacon Room on the second floor. Please RSVP by April 15th. You’re invited if you’re a graduate of our Computer Science, Data Science, or Bioinformatics and Computational Biology graduate program between Summer 2021 through Fall 2022.

Caps and gowns are optional for our departmental event. We recommend dressing nicely if you plan on attending without a cap and gown. Masking is strongly recommended for the event but is no longer required. Light food and refreshments will be provided.

Note this is not the commencement ceremony this is a social gathering for graduates of CS&E programs only. The commencement ceremony will be at 12 pm in Mariucci the same day.

ASE Graduate Commencement Ceremony

ASE Graduate Commencement for 2022 is also Friday, April 29th starting at noon located at 3M Arena at Mariucci. Information detailing the Graduate Commencement event

Spring 2022 Data Science Poster Fair

We invite you to attend the annual Data Science Poster Fair! This year's event will be held on Friday, April 22 from 10:00 a.m. - 12:00 p.m.

Every year, data science M.S. students present their capstone projects during this event as a part of their degree requirements.

The poster fair is open to the public and all interested undergraduate and graduate students, alumni, staff, faculty, and industry professionals are encouraged to attend.

10 am - 11 am - Session 1

Poster 2: Miguel Miguélez Díaz

Advisor: Dongyeop Kang

Capstone: Sentiment Analysis using NLP: Contextual Saliency Variation

Abstract: While huge pre-trained language models have shown a significant improvement on many Natural Language Processing (NLP) tasks, they are often considered as black box models. Thus, there is a strong need of understanding how those models work. This project is focused on the field of sentiment analysis using NLP. Fine-tuned BERT models for sentiment analysis have reported very accurate results to determine if a piece of text shows a positive, neutral, or negative sentiment. Nonetheless, they can be improved and still lack interpretability. In order to understand why a sentence is flagged as positive or negative, the use of integrated gradients (IG) is a powerful tool. Attribution scores for each token in a given sentence are assigned by IG. Depending on the value of the attribution scores, it can be determined if a specific word (token) is conveying a positive or a negative message. The goal of this project is to understand how stable those attribution scores are for each word when some of the surrounding words in the sentence change. After carrying out this analysis, we hope to understand whether this technique is truly useful to interpret sentiment analysis models. Another aim is to detect which words may need more presence in the training step, as well as easing the process of human annotation of attribution scores.

Poster 3: Tiruo Yan

Advisor: Mochen Yang

Capstone: Efficiently Increase Anchor Customers by Scoring and Prioritizing Sysco Leads

Abstract: Due to the impact of the Covid-19, the food-service market became smaller, and the average customer is spending less. Therefore, Ecolab needs to resolve two issue:

•Increase in gaining new customers and

•Growing anchored customers from its existing customers, in order to hit its business goals.

However, Ecolab’s sales organization faces barriers achieving these two goals:

•Limited Capacity: The Ecolab’s sales organization size remains the same despite the need to gain new locations

•Relatively New: Over 50% of the sales representatives in the organization are new to their role.

In the traditional sales process, a large amount of leads were submitted by Sysco, one of Ecolab’s distributors, to their sales representatives. Then the sales representatives will need to perform qualification on the leads sent to them, act on qualified leads and finally turn a portion of it to their customers. This process can have the following potential obstacles:

•Large Supply of Leads: thousands o leads come in from a variety sources, and all of them need to go through qualification steps

•Lead Qualification is Time-Consuming: sales representatives are spending a lot of time gathering information to qualify leads

•No Framework for Actions: there is no framework to help sales representatives to make decisions, action is largely based on judgement

These obstacles then lead to our key question: How do we provide Ecolab’s Sysco sales team with prioritized leads based on their likelihood of purchasing, so they can increase new customers more efficiently? This lead scoring opportunity has two primary objectives:

•How do we gain new customers from the Sysco distributor partnership?

•How do we grow the anchored customer base from existing pool of non-anchored cus-tomers?

To help the sales representative make the lead identification process more efficient, the team focused on building classification machine learning models to prioritize the lead based on the likelihood of purchasing an anchor product.

Poster 4: Qi Le

Advisor: Jie Ding

Capstone: Privacy-Preserving Personalized Recommender Systems with Federated AutoEncoders

Abstract: Two long-standing challenges in Recommender Systems are the data sparsity problem that often arises when users rate very few items and the privacy problem that user's data can be leaked by malicious attacks. Privacy-Preserving Personalized Recommender Systems with Federated AutoEncoders (PPPR) aim to improve the recommendation performance for single user while keeping the data private simultaneously. The existing works assume that the data of different users can be fully shared, and the computation can be performed in a centralized manner. However, the sharing of private data, models, and recommendation tasks is not allowed in many realistic scenarios for the reason that personal data has privacy issue. Moreover, recent attacks demonstrate that simply maintaining data locality during training processes does not provide sufficient privacy guarantees. Existing federated learning approaches either use secure multiparty computation (SMC) which is vulnerable to inference or differential privacy which can lead to huge drop in performance. This work proposes a PPPR based on AutoEncoders and Federated Learning to help single user improve their recommendation performance simultaneously without sharing sensitive assets. Extensive experiments demonstrate that our method significantly outperforms the case where each domain is locally trained, and it performs competitively with the centralized training where all data are shared. As a result, PPPR can effectively integrate large amount of users to form a community of shared interest and provide personalized recommendation for each user.

Poster 5: Sai Sharan Sundar

Advisor: Gregory Pawloski

Capstone: Isolating Supernova Neutrinos using Machine Learning Techniques

Abstract: The NOvA experiment was designed primarily as an accelerator-based oscillation one to detect neutrinos in Fermilab’s NuMI beam. Neutrinos travel between the near and far detectors and the interaction between the neutrinos and the nuclei in the NOvA Cells (a PVC cell filled with a scintillator) is recorded as a hit-map and other variables. The goal of this project is to isolate neutrinos that come from supernova-like events in outer space using machine learning. Previous work included analysis of the hit-map using a convolutional neural network (CNN) but it produced poor result as it contained no additional information about the neutrinos. However, application of a random forest algorithm to the additional variables recorded proved yielded baseline results. Further work is to apply several other models such as Logistic Regression, k Nearest Neighbors and Multi-Layer Perceptron as well explore variable transformations.

Poster 6: Rui Zhou

Advisor: Galin Jones

Capstone: Statistics of Supernova Siblings in Host Galaxies from 1900 through 2022

Abstract: Galaxies similar to the Milky Way are expected to host one supernova (SN) per century. However, some galaxies are more prolific at producing SN siblings than others; more than 2 supernovae (SNe) have been observed in those galaxies within the last century. SNe are classified as type I (Hydrogen-poor) and Type II (Hydrogen-rich). Type I SNe have three sub-classes, type Ia, type Ib, and type Ic. It has been a hot research topic to many about the demographics of SN siblings in galaxies that host multiple events. However, it is hard to have a comprehensive analysis with a small sample size. In our study, we analyze samples from the Open Supernova Catalog. Open Supernova Catalog is a platform that gathers all historical SN discoveries through January 2021. We download and process all SN records from the Open Supernova Catalog (up to the time of 2022/01/11). We classify host galaxies into those that host a single SN and those that host multiple supernovae. After using importance sampling to avoid selection effects present in the surveys, we analyze the distributions of SNe in different groups of galaxies. We utilize Monte Carlo simulations to calculate the expected probability of finding the same type of SNe in consecutive order. Our importance sampling results show that the SN Ia rate is higher in single SN host galaxies (5.1 σ). The SN Ib rate is higher in multiple SNe hosts galaxies (1.5 σ). Our Monte Carlo simulation suggests that SN Ib is more likely to occur consecutively (1.8 σ). In conclusion, the patterns of seeing a higher rate of SN Ia in single SN hosts and a higher rate of SN Ib in multiple CC hosts are partly due to selection effects and the other factors that might include the intrinsic properties of their host galaxies.

11 am - 12 pm - Session 2

Poster 8: Rebecca Dura

Advisor: Gregory Pawloski

Capstone: Identifying Neutrinos Associated with Supernova Events

Abstract: With the rise of multi-messenger astrophysics in recent years, astronomers and data scientists have begun using data from multiple messengers, including light, particles, and gravitational waves, to reveal more information about the Universe than ever before. This project explores the relationship between neutrino particles that are detected by Fermilab's detectors and supernova events. The main goal of this research was to build a machine learning model to accurately predict whether a detected neutrino was associated with a supernova event. Multiple approaches were taken with different datasets. The first approach involved passing temporal data through CNNs (convolutional neural networks). The input data for the CNNs was sequential images encoding the neutrinos' energy over time. However, the results revealed that not enough information about the neutrinos was being included in that model. The second approach included much more information about the neutrinos and involved a variety of supervised machine learning models. This approach was much more successful, especially after utilizing variable selection techniques to reduce the dataset size, transforming some of the variables, and applying dimensionality reduction techniques. While supernova events are not frequent, models such as those built with this project will be vital in helping astronomers detect supernova events in the future.

Poster 10: Yue Liang

Advisor: Chih-Lin Chi

Capstone: Patient Centered Prescription using Counterfactual Prediction Optimization Algorithms for Statin Users

Abstract: Clinical decision support (CDS) that does not optimize individual variability with best practice and treatment outcomes, cannot support patient-centered care. For example, a statin treatment CDS focused on cholesterol reduction without accounting for patient comorbidities, at risk lifestyle, or genetics, may well suggest a treatment that increases adverse effects, has a high cost, and has a low potential for improved outcome. We present a novel patient-centered decision support (PCDS) system for statin initiation that accounts for best practice and patient variation and ranks treatment options by their predicted ability to improve cardiac outcomes and reduce adverse effects for individual patients. A total of 30 M de-identified claim records from 1.4 M adults receiving statin treatment over 1990-2020 are available in the OptumLabs® Data Warehouse and were used to train our counterfactual prediction algorithm. For a new patient, the PCDS system predicts LDL reduction, cardiac events, and other outcomes at 1 and 5 years, for every potential statin treatment. The PCDS then integrates doctor and patient preferences into a multi-objective optimization algorithm to generate a ranked list of treatment options. Our novel patient-centered decision support (PCDS) adds sophisticated AI modeling and optimization to account for evidenced-based best practice, patient variability, and adverse effect risk to produce a clear ranking of treatment options that can be re-ranked upon input of patient preferences and provider’s lifestyle, socioeconomics, culture, experience and environment factors. This PCDS approach leverages traditional CDS adjusted to individual preferences to aligned with the patient-centered care model.

Poster 11: Samuel Johnson

Advisor: Galin Jones

Capstone: Analysis of the Demographics of Supernovae within Host Galaxies

Abstract: Galaxies similar to the Milky Way host supernovae at an average rate of approximately one supernova per hundred years. However, some galaxies haven’t hosted any observed supernova in the past century while others have hosted as many as ten. Supernovae are classified as Type I or Type II based on the presence of a hydrogen emission line in their optical spectra. Supernovae lacking hydrogen are classified as Type I; others are classified as Type II. J.P. Anderson and M. Soto's 2013 paper demonstrated that galaxies tend to host supernovae of the same type. As a result, the demographics of supernovae in galaxies that host multiple supernovae differ from the demographics of supernovae in galaxies that host a single supernova. Using additional supernovae samples from the Open Supernovae Catalog, we rework this analysis by examining the distributions of supernovae types within galaxies that host multiple supernovae and comparing these distributions to the broader distribution of supernovae types across all galaxies. We also study the spectra of galaxies that have hosted multiple supernovae in an attempt to determine the physical cause of the galaxies' tendency to host supernovae of the same type. We find that Type Ib supernovae occur consecutively in host galaxies more than what would be expected based on the overall distribution of supernovae types. We also find that the metallicity of these host galaxies tends to fall in a very narrow range, indicating that a galaxy's metallicity has an effect on the type of supernovae it hosts.

Poster 12: Rosalind Hong

Advisor: Jeff Calder

Capstone: Graph learning techniques in hyper spectral image classification

Abstract: Hyperspectral images are images with many spectral bands, instead of just 3 bands (RGB values) in traditional images. These images are often hard to obtain. Learning with scarce data is never easy, and how to learn effectively with few training samples is an important topic in machine learning literature. Graph learning is a machine learning algorithm that solves discretized PDEs on graphs. With graph learning, reasonably good test accuracy can often be achieved, even when traditional methods fail completely. We apply graph learning techniques to the problem of classifying hyperspectral images and show improvements of test accuracy compared with traditional methods.

Poster 13: Bob Sturm

Advisor: Lucy Fortson

Capstone: Towards Efficient IACT Calibration with Deep learning-based Muon Identification

Abstract: Extensive air showers are initiated when very energetic particles, including very-high-energy gamma rays, enter the earth’s atmosphere. Imaging Atmospheric Cherenkov Telescopes (IACTs) are used to measure the Cherenkov light from charged particles emitted during air showers. Muons are secondary particles emitted during these air showers and IACTs are calibrated by comparing the Cherenkov light signal in IACT images of muons to known theoretically expected values.

The main goal of this project is to improve the efficiency of IACT calibration. With this aim, we a) Implemented a Convolutional Neural Network (CNN) proposed by Flanagan and trained it on a sample of ~600,000 IACT images, b) Explored ways to simplify the model’s complexity and computational performance without loss of prediction performance, c) Developed a muon fitting algorithm that determines a muon ring’s centroid and radius, and d) Analyzed the relationship between muon ring radii and the total electromagnetic signal in images.

Poster 14: Xianjian Xie

Advisor: Jie Ding

Capstone: Client selection in federated learning with adversarial clients

Abstract: Federated Learning(FL) is a machine learning paradigm where many distributed FL clients collaboratively train models without sharing their local data to the FL server. Its characteristics of guaranteeing training quality and data privacy at the same time have attracted more and more attention from industry and academia. In many FL frameworks, the assumption that clients are trustworthy is often made to simplify the problem. However, in practical situations, there may be adversarial clients in the FL network. The model parameters they upload, if adopted, will make final model performance worse. For this reason, client selection strategies play an important role in the model training process. In this paper, we propose a new method to identify adversarial clients early in the training process and remove them from future training.

Please contact Allison Small at csgradmn@umn.edu with any questions.

Last day to cancel full semester classes without college approval and receive a "W"

The last day to cancel full semester classes without college approval and receive a "W" is Monday, March 28.

View the full academic schedule on One Stop.

Main Office	Student Services
4-192 Keller Hall 200 Union Street SE Minneapolis, MN 55455 (612) 625-4002 csdesk@umn.edu	324 Lind Hall 207 Church Street SE Minneapolis, MN 55455 (612) 625-4002 csdesk@umn.edu