What is StemForAll?

StemForAll is a research apprenticeship program for undergraduate students and especially talented high school students in the Rochester, New York, area. The program is built on a simple principle: every STEM student should have an opportunity to learn how research is actually done.

Participants work in small groups with faculty members, graduate students, and other researchers on open-ended problems in mathematics, data science, physics, biology, and related fields. The purpose is not merely to learn established material. Students learn how to formulate questions, read technical literature, test ideas, use computation effectively, communicate mathematics and science, and persist when a problem does not have an immediate answer.

How the program works

Spring preparation Students begin individual preparation before the summer program through readings, preliminary exercises, programming, and conversations with project supervisors.

Summer intensive The central two-week workshop brings all research groups together for daily project work, mini-courses, cross-group interaction, and final presentations.

Fall and winter continuation Promising projects may continue after the workshop, allowing students to deepen their work and, in some cases, contribute to presentations, software, or research papers.

Program history and growth

The present Rochester program has been running in its StemForAll form since 2018. Earlier versions of the undergraduate research program were organized at the University of Rochester and the University of Missouri beginning in 2001. Many former participants have gone on to graduate study, Ph.D. programs, research careers, and other STEM professions.

The broader goal is to develop a model that can be reproduced at universities throughout the country. In addition to Rochester, the program has now been implemented at Penn State University, and programs are planned at Missouri State University and Virginia Tech. Each site can adapt the model to its own faculty strengths while preserving the central commitment to local, inclusive, sustained research training.

Links to previous years are available on the program archive page.

StemForAll 2026 at a glance

Eight research projects Topics include signal recovery, fractal analysis of EEG data, topology, geometric PDEs, neural networks, financial mathematics, and quantum systems.

More than twenty instructors and supervisors The instructional team represents the University of Rochester and numerous partner institutions.

A regional and national student community Participants include students from the University of Rochester, RIT, SUNY Geneseo, NYU, Baruch College, and local high schools, as well as students joining from outside the region.

Participation

All interested Rochester-area undergraduate students are welcome to register. Advanced high school students who are prepared for sustained mathematical or scientific work are also encouraged to participate. Registration is used to match students with suitable projects; it is not intended to restrict the program to students who already have research experience.

The program is designed to welcome students with different levels of preparation. Each project description below explains the relevant background, and supervisors provide preparatory material when specialized knowledge is needed.

Partnership and sponsorship

StemForAll is seeking partners who share the goal of expanding access to authentic STEM research. Corporate and philanthropic support can help provide student transportation and meals, computing resources, graduate-student and junior-mentor support, project materials, research showcases, and the development of a practical guide for universities that wish to establish similar local programs.

Organizations interested in supporting StemForAll, providing mentors, or helping the program expand to additional universities are invited to contact Alex Iosevich through the University of Rochester Department of Mathematics.

Program instructors and research projects

Program instructors: Karam Aldahleh (Oxford), Jacob Canel  (Penn State), Jiajun Du (UR), Sam Ferguson (NYU), Vishal Gupta (UR), Gabe Hart (UR), Alex Iosevich (UR), Joshua Iosevich (RIT), Alhussein Khalil (MIT), Steven Kleene (UR), Neeraja Kulkarni (UR), Tran Duy Anh Le (UCSD), Alex Liu (UR), Azita Mayeli (CUNY), Svetlana Pack (Penn State), Alice Quillen (UR), Akshay Sant (UR), Steven Senger (Missouri State), Liz Tatum (UR), John Werren (UR), Kunko Zou (UR), and Yan Zou (UR) Structure of the workshop: StemForAll2026 is going to consist of a wide variety of projects, listed below, involving the interaction of pure and applied mathematics, statistics, physics and data science. Many of the projects are strongly related, which should lead to a considerable amount of interaction. We are going to have a blast!

StemForAll2026 Projects: There are 8 projects running this year. Projects T1, T2, T3 and T4 are theoretical projects, though each has an applied component. Projects A1, A2 and A3 are applied, though each has a theoretical component. Project P1 is a physics project related to a number of ideas pursued by the other research groups.

In addition, three mini-courses will take place in Landers Auditorium during the first week of the program: Signal recovery (A. Iosevich), Python programming (to be determined), and Representation theory with applications to signal recovery (K. Aldahleh).

(T1) Signal Recovery: The basic problem in signal recovery is to send a finite signal via its Fourier transform, with some of the Fourier values missing, and recover the signal exactly if a reasonable set of assumptions is satisfied. In practical situations, the values of the original signal are missing. The assumptions are on the Fourier coefficients, and the recovery is approximate, not exact. This has significant applications in industrial data science. In this project we are going to explore both pure and applied aspects of signal recovery. The applied and pure directions will be explored together, with significant daily interactions between the research groups.

This year, we are going to run three research subgroups on Signal Recovery. The non-abelian recovery group, run by Karam Aldahleh, Alex Iosevich, Azita Mayeli, Alhussein Khalil, and Akshay Sant is going to explore signal recovery in the finite non-commutative setting (NC) where representation theory is going to play a key role. The Fourier ratio and graph recovery group (G), run by Sam Ferguson, Vishal Gupta, Alex Iosevich, Joshua Iosevich, is going to investigate recovery problems on graphs. The Parseval frames recovery group (P), run by Alex Iosevich, Tran Duy Anh Le, and Azita Mayeli is going to explore uncertainty principles and signal recovery in the context of Parseval frames. The three subgroups will meet in Hylan 201 and Hylan 202.

Project supervisors: Karam Aldehleh, Sam Ferguson, Vishal Gupta, Gabe Hart, Alex Iosevich, Joshua Iosevich, Azita Mayeli, Akshay Sant, and Tran Duy Anh Le

Participants: Gideon Afriyie G1 (UR), Karam Aldahleh NC1 (UR), Kenneth Fei NC2 (UR), Gabe Hoag NC3 (UR), Suleman Khan P1 (Baruch), Misha Jindal P2 (UR), Allhussein Khalil NC4 (UR), Julian King NC4 (Geneseo), Austin O'Connor P3 (UR), James-Lucius Okenwa G2 (UR), Gus Smith G3 (UR), Ben Song G4 (Pittsford High School), Suva Parvin Srithe G5 (UR), Erdem Togay P4 (UR), Sinan Yang G6 (UR), Fiona Zhang G7 (UR), Yiyi Zhao P5 (UR)

Location: Hylan 201, Hylan 202, and Hylan 909

Background reading: Will be continually made available on the Discord channel.

(T2) Analyzing the Fractal Dimension Distribution of EEG Data to Classify Seizures: Real world data is replete with fractal behavior, highly geometrically complex patterns that- in a meaningful sense- have a fractional  dimension. These patterns are, by nature, hard to approximate in finite data and as such are underexploited by standard data analysis techniques. However, it is possible to learn a lot about the distribution underpinning data using fractal analysis - for example,  past research has indicated the electroencephalogram (EEG) recordings of brain waves have different fractal dimensions when in an epileptic vs. nonepileptic regime, suggesting that measuring fractal behavior could be useful for detecting seizure activity in EEG data.

We plan to approach this problem with the following goals:

i) Develop more sensitive mathematical tools for estimating the fractal dimension distribution of real world time series data
ii) Apply these tools to identify seizures by detecting changes in the fractal dimension distribution over a sliding time window, and
iii) Compare our methods to the industry standard multi-fractal analysis methods.

Project supervisors: Svetlana Pack and Jacob Canel

Participants: Grace Brandt (UR), Josie Elliston (UR), Oscar Jackson (UR), Aashi Jindal (unaffiliated), Ethan Kang (UR), Nadia Lach-Hab (UR), Ziqing Li (UR), Shana Kester (UR), Pitambar Pandey (UR), Farid Rohan (UR), Claire Strobel (UR), Ria Vaish (UR), Jin Yuan (UR), Lior Zendel (UR)

Location: Hylan 203

Background reading: The definitions of fractal dimension and dimension distribution that we are working with are expressed in the language of measure theory. We do not expect anyone to come in with any measure theory background, however we did prepare a document that includes a crash course in all the measure theory and fractal analysis you need to understand the motivation behind our fractal analysis technique. We highly encourage you to look through at least Section 1 of this document over the coming weeks if you have time.

If you want to do even more supplemental reading to get comfortable with the notion of fractal dimension, Hausdorff dimension, that we will talk about, we suggest you to read sections  3.1, 3.2, 3.3 in Falconer. These materials will be available on Discord .pdf format.

(T3) Fixed Points of the Conjugate in the Steenrod Algebra: The Steenrod algebra is an algebraic object with many interesting connections to topology. The Steenrod algebra comes with a great deal of intricate structure: it has its own versions of addition and multiplication, as well as more novel operations. In particular, the Steenrod algebra comes equipped with an automorphism called the conjugate, that is, a special function from the Steenrod algebra to itself.

A more familiar example of an algebraic object with a simpler conjugate is the complex numbers: the complex numbers come equipped with addition and multiplication, as well as the complex conjugate. The fixed points of the complex conjugate are the real numbers, because the complex conjugate of any real number is itself.

Since the Steenrod algebra has a more complicated structure than the complex numbers, the fixed points of its conjugate are not known. The goal of this project is to investigate and identify the fixed points in small portions of the Steenrod algebra. This is a good project for students interested in algebra, combinatorics, or topology, and no specific background is required.

Project supervisor: Elizabeth Tatum

Participants: Ishika Jahaly (U of Mauritius), Logan Singh (UR), Showmee Zhou (UR)

Location: Hylan 206

Background reading: Will be continually made available on the Discord channel.

(T4) Numerical solutions of PDEs arising in geometry: Mean curvature flow is a second order geometric evolution equation on surfaces that  can be thought of as the negative gradient flow for the area functional. To decrease the value of a function most efficiently, one moves in the direction of the negative gradient of the function. Thus, to decrease the area of a surface most efficiently, one deforms it according to the mean curvature flow. The process is non-linear and exists for short time scales and almost always develops singularities in the surface, which are modeled on  "self shrinkers", surfaces which shrink under the flow. Exact solutions to the flow, especially embedded ones, are of central interest to the field.  In this project, the group will numerically approximate a family of  immersed--the surfaces self-intersect--rotationally symmetric  self shrinkers for the mean curvature flow in Euclidean three space  constructed by Stephen J. Kleene and Niels Martin Moller in 2010. If it can be shown that the members of the family are non-degenerate, they can be desingularized and new families of embedded self shrinkers can be shown to exist.

Project supervisor: Stephen Kleene

Participants: Evan Balder (RIT), Yoonseo Han (UR), Zoe Juergensen (UR), Braden Lenn (UR)

Location: Hylan 1011

Background reading: Will be continually made available on the Discord channel.

(A1) Sales modeling: We are going to build and test neural network models with economic indicator regressors to effectively predict future sales in retail. A variety of neural network models will be built using TensorFlow, Keras, facebook prophet and others. The second major part of this project is using generative AI for parameter tuning and inventory optimization. Theoretical aspects of this problem will be considered as well. Connections with the theory of exact signal recovery will also be explored.

Project supervisors: Vishal Gupta, Gabe Hart, and Alex Iosevich

Location: Hylan 1106A

Participants: John Anastasi (UR), Yinuo Feng (UR), Jun Hung (UR), Arjun Kanani (UR), Naomi Kim (UR), Michael Koyfman (high school), Bowen Li (UR), Linh Nguyen (UR), Kunxu Song (UR), Anh Tran (UR), Zesheng Yu (UR), Emily Yun (UR), Yining Zha (UR), Yichen Zhang (UR), Owen Zhao (UR)
Background reading: Will be continually made available on the Discord channel.


(A2) Useless neurons: Not every neuron in an artificial neural network contributes equally to its performance. In fact, many neurons can be removed—or pruned—with little or no loss in accuracy. When done carefully, pruning can greatly reduce the computational cost of the inference stage, making models faster and more efficient. For this reason, a variety of methods have been developed to identify neurons that are relatively unimportant or redundant. In this group, we will use ResNet-18 (and possibly ResNet-50) as a case study to evaluate the effectiveness of different pruning techniques. We will also examine the theoretical ideas behind these methods and explore the possibility of developing new pruning criteria.

Familiarity with Python and multivariable calculus is preferred but not strictly required.

Project supervisor: Kunko Zou and Yan Zou

Participants: Wenzan Fan (UR), Jiarui Feng (UR), Sanjay Makam (RIT), Alex Phillips (UR), Armaan Sapra (UR), Zihao Wang (UR), Zhongtian Zhai (UR), Daiming Zhou (UR), Kunko Zou, Yan Zou

Location: Hylan 1106B

Background reading: Will be continually made available on the Discord channel.

(A3) Financial Mathematics: Real-Time “Hot Search” Topic Rankings for Predictive Asset Pricing and Event-Driven RL Financial Trading: This project examines whether platform-curated hot-search / trending topic rankings generate short-horizon attention and information shocks that predict returns, volatility, and liquidity. An end-to-end, reproducible pipeline is developed to (i) continuously ingest ranked hot topics via APIs/snapshots, (ii) map topics to tradable objects (stocks/industries/indices) via entity linking and candidate ranking, (iii) infer structured semantics (direction, horizon, confidence, type) using an NLP pipeline, and (iv) train a risk-aware, event-driven RL trading policy (SMDP/event-triggered) under realistic constraints and transaction costs. Expected outcomes include rigorous event-study and cross-sectional tests, a leakage-safe evaluation framework with open-source code, and a real-time monitoring dashboard that produces time-stamped topic-to-asset signals for paper trading and research auditability.

Python, probability, linear algebra, time series, and asset pricing are strongly recommended; familiarity with NLP, ML, and RL is preferred but not required.

Project supervisor: Jiajun Du

Participants: Polina Chub (UR), Xingyan Jin (UR), Hanzhang Li (UR), Jingyi Li (UR), Ilia Lukinov (UR)

Location: Hylan 1101

Background reading: Will be continually made available on the Discord channel.

(P1) Quantum random walks and quantum random circuits: A quantum walk is a quantum mechanical equivalent of a classical random walk. A popular type of quantum random walk involves discrete and iterated local unitary transformations for quantum states that are connected by a graph. In contrast, quantum random circuits are composed of a series of local unitary transformations that are sampled independently according to the Haar probability measure. Quantum walks and random circuits have many applications including in quantum computing, quantum simulation, condensed matter physics and for demonstrating quantum advantage.

We will explore the properties of more general classes of iterated quantum transformations on quantum states connected by a graph that include both random and fixed unitary components and possibly classical components such as state initialization. Much of our exploration is likely to involve calculations in Python, though we will also try to gain understanding analytically.  Project supervisor: Alice Quillen

Participants: Benjamin Akararatovic (UR), Nicolas Cannella (NYU)

Location: Hylan 1104

Background reading: Will be continually made available on the Discord channel.