PIMS Mitacs--India Globalink Student Landing Page

• Indian Institute of Science (IISc)
• Indian Institute of Technology (IIT), Bombay
• Indian Institute of Technology (IIT), Delhi
• Indian Institute of Technology (IIT), Kanpur
• Indian Institute of Technology (IIT), Kharagpur
• Indian Institute of Technology (IIT), Madras
• Vellore Institute of Technology (VIT)
• Indian Institute of Technology (IIT), Roorkee
• Indian Institute of Technology (IIT), Hyderabad
• Indian Institute of Technology (IIT), Hyderabad
• Indian Institute of Technology (IIT), Guwahati
• Indian Institute of Technology (IIT), BHU-Varansi
• Anna University
• SRM Institute of Science and Technology (SRMIST)
• Manipal Academic of Higher Education (Manipal Institute of Technology)
• BITS Pilani
• Amrita Vishwa Vidhyapeetham
• Kalinga Institute of Industrial Technology (KIIT)
• Thapar Institute of Engineering and Technology
• Indian Statistical Institute
• National Institute of Technology Tiruchirappalli
• Chennai Math Institute (CMI)

Department: Mathematics and Statistics
Preferred background: Mathematics
Preferred start date: July 5, 2027 (flexible)

Topology is the mathematical study of spaces. Professor Frankland's research area is algebraic topology, which describes the shape of spaces using algebraic quantities that can be computed — for example, telling apart a sphere from a torus using the fundamental group or homology groups. His work looks at the extent to which such algebraic invariants determine a space, and how to put as much structure as possible on them to retain more information.

Representation theory studies groups by examining how they act on other objects. A representation of a group G over a field k is a k-vector space on which G acts — equivalently, a (left) module over the group algebra kG. When the characteristic of k divides the order of G, kG-modules behave very differently, which is the subject of modular representation theory. This project focuses on the case where k has characteristic p and G is a p-group (of order a power of p).

The kG-modules that are not projective are of particular interest. The stable module category StMod(kG) is defined by quotienting out homomorphisms that factor through a projective module — an example of a triangulated category. The goal of the project is to write code to carry out computations in stable module categories. The motivation is that stable module categories provide a good testing ground for Professor Frankland's broader work on triangulated categories.

What you'll bring:

On the theoretical side:

• A course in linear algebra
• A course in abstract algebra (e.g., group theory, rings, fields, and modules)

On the practical side:

• Some coding experience, in any programming language
• Ideally, some experience with mathematical software such as GAP, Sage, Mathematica, or Maple

Department: Mathematics and Statistics
Preferred background: Mathematics
Preferred start date: July 5, 2027 (flexible)

Professor Frankland's research area is algebraic topology, which describes the shape of spaces using algebraic quantities that can be computed. His work looks at the extent to which such algebraic invariants determine a space, and how to put as much structure as possible on them to retain more information.

Differential graded algebras (dg-algebras) are important both in algebra and in topology. Given a dg-algebra A, one often wants to compute its homology H_*A. If the dg-algebra is known explicitly, computing the homology becomes, in principle, a problem in linear algebra — but a dg-algebra is not always described in terms of explicit vector spaces and linear transformations.

The goal of this project is to write code to compute the homology of a dg-algebra described by a presentation — that is, a list of generators and relations. The motivation is to compute the homology of dg-algebras that carry universal examples of Massey products in their homology; solving this would shed light on the structure of Massey products in general.

What you'll bring:

• A course in linear algebra
• A course in abstract algebra (e.g., group theory, rings, fields, and modules, commutative algebra, or homological algebra)
• Ideally, some experience with mathematical software such as Sage, GAP, Mathematica, or Maple

Department: Civil, Geological, and Environmental Engineering
Preferred background: Engineering, mathematics, statistics, computer science
Preferred start date: May 1, 2027 (flexible)

Dr. Abdelrahman's research focuses on developing sustainable, resilient, and durable structures through experimental testing, high-fidelity computer modelling, and machine-learning-enabled structural assessment under extreme loading and varying climate conditions.

Masonry structures make up a significant share of the building stock in both Canada and India, two countries that face substantial seismic hazard, and masonry walls are particularly vulnerable to earthquake damage. Rapid, reliable damage assessment is essential for post-earthquake recovery and decision-making, but conventional visual assessment can be slow and subjective. Advances in artificial intelligence and computer vision offer an opportunity to build accessible tools for automated damage assessment.

This project gives senior undergraduates a focused research experience at the intersection of artificial intelligence, mathematical modelling, and earthquake engineering. Interns will co-develop and evaluate an AI-based model for identifying and classifying seismic damage in masonry walls from images.

The work involves assembling and preprocessing a dataset of masonry-wall images from laboratory experiments or public sources, identifying damage features such as cracks, crushing, and separation, and training a machine-learning model to classify damage levels. Interns will investigate how model performance is affected by the size and diversity of the training data, and explore ways to improve model reliability.

The project offers hands-on experience in data preparation, machine learning, mathematical analysis, and structural engineering, working closely with a USask faculty member and gaining exposure to interdisciplinary research practice.

A key outcome will be a preliminary, reproducible AI framework for masonry damage assessment, along with an evaluation of its accuracy and potential for further development — a strong foundation for pursuing graduate studies in mathematical sciences, artificial intelligence, or structural/earthquake engineering.

What you'll bring (some of the following are preferred, not all required):

• Basic experience with programming/data-analysis tools such as Python or MATLAB
• Basic understanding of machine-learning concepts: classification, training/testing datasets, model evaluation, overfitting
• Basic knowledge of statistics, probability, linear algebra, or numerical methods
• Ability to interpret experimental or numerical results and relate damage patterns to basic mechanics concepts
• Ability to summarize results, prepare figures and presentations, and contribute to technical reports

Department: Mathematics
Preferred background: Mathematics, computer science
Preferred start date: May 1, 2027 (flexible)

Dr. Kundu's research interests include Diophantine equations, Iwasawa theory, and Beyond Endoscopy. Her PhD work studied growth patterns of fine Selmer groups in infinite towers of number fields, and she has also worked in non-commutative Iwasawa theory, arithmetic statistics, Diophantine stability, Hilbert's 10th Problem, and the Langlands program.

This project investigates how the singular value decomposition (SVD) can be used to study structural features of digital images, with particular emphasis on artistic images. A grayscale image can be represented by a matrix, and the SVD expresses that matrix as a sum of rank-one components weighted by singular values; the rate of decay of those singular values measures how efficiently an image can be approximated by a low-rank matrix.

The first part of the project develops the necessary mathematical background — matrix rank, orthogonality, projections, matrix norms, the SVD, and the Eckart–Young theorem on best low-rank approximation. The intern will then explore how these ideas behave across different classes of images and formulate hypotheses about the relationship between singular-value decay and visual complexity.

A central research question is whether singular-value-based quantities can distinguish broad types of visual structure — for example, comparing geometric abstract paintings with highly textured works to see whether their singular values decay at systematically different rates. Possible numerical invariants include effective rank and the number of singular values needed to capture a prescribed percentage of total matrix energy.

The project will also probe the limits of these invariants, searching for visually simple images with unexpectedly high rank (or visually complex images with efficient low-rank approximations), and studying stability under resizing, cropping, or noise. It combines theoretical linear algebra with computational experimentation, hypothesis formation, and data analysis, culminating in a written report.

What you'll bring:

• Linear algebra
• Familiarity with a programming language (Python, C, or C++)
• An open mind

Department: Computer Science
Preferred background: Mathematics, computer science, engineering, physics
Preferred start date: May 3, 2027 (flexible)

Professor Spiteri's research spans numerical analysis, scientific computing, and high-performance computing, with a focus on algorithms and software for AI-enabled scientific computing, emerging heterogeneous computing platforms, and the actor model of concurrent computation as a new paradigm for large-scale scientific simulation. Current projects include real-time simulation of cardiac electrical activity, molecular dynamics for drug design, large-scale hydrological flow simulation, and plasma simulation — aimed at advances in personalized medicine, water security, flood prediction, and clean (fusion) energy.

This project develops cutting-edge algorithms and software for large-scale simulations of complex systems, typically governed by nonlinear partial differential equations. Interns can choose from one of four directions, matched to their interests:

1. Cardiac electrophysiology: the extracellular–membrane–intracellular (EMI) model offers unprecedented physiological detail compared to standard models. Much work remains to determine when this added detail yields meaningful predictive advantages, especially in personalized medicine and diseases that current models cannot adequately represent.
2. Drug and vaccine design: this complex, multi-stage process often requires substantial human intervention and uses computing resources inefficiently. This direction explores using the actor model of concurrent computation to automate and streamline the workflow, accelerate discovery, and expand the range of tractable design problems.
3. Hydrological flow simulation: current numerical methods for continental- to planetary-scale hydrological flow are often inefficient and insufficiently robust. This direction emphasizes time integration, particularly operator-splitting methods, to improve the efficiency, reliability, and fidelity of hydrological software.
4. Plasma simulation: central to both long-range space propulsion (e.g., Hall thrusters) and fusion energy, these simulations are highly challenging, and accurate, efficient software often decides whether results are reliable or spurious. This direction focuses on plasma simulation on specialized hardware, including GPUs and Apple silicon.

Across all four directions, interns may also engage with new approaches to heterogeneous computing developed in the lab, combining mathematical modelling, numerical analysis, scientific software development, and high-performance computing.

What you'll bring:

A strong background in applied mathematics, computer science, or a related area such as engineering or physics, together with some experience in numerical analysis, scientific computing, and computer programming. Depending on the chosen direction, familiarity with high-performance computing, parallel or distributed computing, machine learning, chemistry, or electrophysiology would be beneficial.

You should be comfortable implementing and running software, processing and analyzing results, and presenting findings orally and in writing. Experience designing computational experiments, comparing numerical approaches, or developing and testing algorithms and software is valuable. Because this is a highly multidisciplinary project, strong communication, teamwork, and problem-solving skills matter as much as technical ability — curiosity, initiative, and a willingness to ask questions and chart new directions are especially welcome, since some aspects of this work (particularly actor-based scientific computing) remain relatively unexplored.

Department: Physics & Astronomy
Preferred background: Physics; engineering physics; computing; statistics; data science
Preferred start date: May 1, 2027 (flexible)

Professor Das is a theoretical physicist working where quantum mechanics, gravity, and cosmology meet. His group studies how quantum gravity could be detected in laboratory and astrophysical data (quantum gravity phenomenology, including the Generalized Uncertainty Principle), how quantum effects and modified gravity can remove the Big Bang and black hole singularities, and the origin of dark matter and dark energy — combining analytic calculations with numerical simulation and statistical analysis of observational data.

Standard cosmology traces the universe back to a Big Bang singularity, where Einstein's equations break down. Quantum-gravity corrections and modified theories of gravity can replace this singularity with a "bounce," in which an earlier contracting phase turns into today's expansion. Professor Das's group has constructed several such models, including power-law bounces in f(R) gravity and quantum-corrected Friedmann equations; which of them survive current data is an open question.

The intern will build a pipeline that tests these models against observations:

1. Weeks 1–2: learn the essentials — Friedmann equations, cosmological distances, Bayesian inference — through guided reading and exercises.
2. Weeks 3–5: implement the expansion history of two or three bouncing models by solving their equations numerically, and check them against the standard ΛCDM model.
3. Weeks 6–9: fit the models to public data (Pantheon+ supernovae, DESI baryon acoustic oscillations, cosmic-chronometer measurements, and Big Bang nucleosynthesis constraints) using Markov chain Monte Carlo and nested sampling.
4. Weeks 10–12: compare the models with ΛCDM using information criteria and Bayesian evidence, identify which parameters the data constrain, and present the results in a report and talk.

Expected outcomes are constraints on the bounce parameters, a ranking of the models against ΛCDM, and documented, reusable code. The intern will receive regular one-on-one supervision, work alongside graduate students, and attend departmental seminars.

What you'll bring:

Essential:

• A solid foundation in calculus, ordinary differential equations, and linear algebra
• Introductory classical mechanics, quantum mechanics, general relativity, and basic probability and statistics
• Working knowledge of Python (or a comparable language), and a willingness to write clear, documented code
• Ability to read technical material independently and communicate results clearly in English

Desirable (not required):

• Coursework or self-study in special or general relativity, cosmology, or quantum mechanics
• Exposure to statistical inference, curve fitting, or Monte Carlo methods
• Experience with Git, Jupyter notebooks, and LaTeX

Prior knowledge of cosmology is not required — the first two weeks provide structured training. The project suits a motivated senior undergraduate in physics, engineering physics, mathematics, or computer science who is curious about fundamental questions and enjoys combining theory with computation.

Department: Mathematics and Statistics
Preferred background: Mathematics, computer science, statistics, physics
Preferred start date: [CHECK WITH SUPERVISOR — spreadsheet currently shows May 1, 2025, which is almost certainly a typo for 2027]

Professor Ray's research uses techniques from probability theory to understand objects arising in statistical mechanics and mathematical physics, with a current focus on random trees, random height models, and disordered models.

Take a graph G = (V, E) and a function f from the vertex set to the integers such that adjacent vertices differ by at most 1. Fix a boundary set B and a compatible set of allowed values on B (usually 0), and uniformly pick a function matching those boundary values — this is a uniform integer-valued Lipschitz function. Its general theory is intricate and depends heavily on the underlying graph.

This project studies such functions on a regular tree of finite height under various boundary conditions. On a tree, the distribution of the root value satisfies a recursive distributional equation. With the 0 boundary condition, Professor Ray and co-authors proved that the root distribution converges when the tree's branching number d is strictly smaller than 8; with other boundary conditions (say ±10), much less is known — it isn't even clear whether the root distribution is unimodal or bimodal. The project involves analyzing the recursion computationally, forming conjectures about how the root distribution depends on the boundary conditions, and exploring related recursive distributional equations.

What you'll bring (desired, not essential):

• A good mix of probability theory, analysis, and combinatorics
• Basic coding skills, e.g., Python (this requirement may be eased with the help of AI tools)
• Some basic proof-writing skills

Department: Mathematics and Statistics
Preferred background: Mathematics, physics, computer science, statistics
Preferred start date: May 1, 2027 (flexible)

Professor Ray's research uses techniques from probability theory to understand objects arising in statistical mechanics and mathematical physics, with a current focus on random trees, random height models, and disordered models.

This is a fairly open-ended project on disordered models. Let G be a finite graph, and consider the space of subgraphs with some type disallowed: forbid cycles and you get a forest; forbid vertices of degree two or more and you get a matching; require every vertex to have degree two with the graph connected and you get a Hamiltonian cycle. Now place random, independent weights on the edges — what can be said about the constrained subgraph that maximizes this weight? This constrained optimization problem, with no clean method of attack, is closely related to the spin-glass model.

The project explores variations of these models and recent and classical results, working toward formalizing them. The major open problems in this area are hard, but Professor Ray has several tractable "toy" problems on trees suited to an internship — a good entryway into the field, with the student reading relevant papers, developing intuition for the area, and tackling closely related toy open problems.

What you'll bring (desired, not required):

• Good exposure to probability theory
• A general inclination toward graph-theoretic problems (no prior graph theory knowledge required)

Department: Electrical and Computer Engineering
Preferred background: Computer science; mathematics; mathematics and computing; electrical/electronics and computer engineering
Preferred start date: May 17, 2027 (flexible)

The Ko Lab works on computer architecture and arithmetic, and on efficient hardware for compute-intensive workloads, including cryptographic circuits and post-quantum cryptosystems for resource-constrained devices. This project sits where that meets lattice-based standards such as ML-KEM (Kyber) and ML-DSA (Dilithium), the lattice-reduction algorithms (LLL, BKZ) used to attack them, and the cost models that translate attack performance into the parameter sizes hardware must implement.

Canada is moving its systems from RSA and elliptic-curve cryptography to the lattice-based post-quantum standards ML-KEM and ML-DSA, with a federal roadmap targeting 2035 for completing the migration. The parameters of these standards are chosen so the best known lattice attacks are infeasible, yet the cost of those attacks is predicted by analytic models — chiefly the Lattice Estimator — rather than measured directly. This 12-week computational project lets interns test those predictions at sizes where attacks still succeed.

Interns will implement textbook Regev-style LWE encryption and textbook NTRU in Python, generate seeded instances of increasing dimension, modulus, and noise width, and attack them with LLL and BKZ lattice reduction in fpylll — using the primal embedding attack on LWE and the Coppersmith–Shamir lattice on NTRU. For every instance, they will record attack success, the smallest successful BKZ block size, running time, reduced-basis quality, and the floating-point precision required, comparing these measurements with the Lattice Estimator, the 2016 success estimate, and the fpylll BKZ simulator, then extrapolating to standardized ML-KEM and ML-DSA parameter sets.

Deliverables include an open-source code repository, a reproducible dataset, analysis notebooks showing where attacks stop succeeding and how closely the models track that boundary, a note on how each parameter contributes to security, a final report and presentation, and a reusable teaching module.

The project supports the post-quantum cryptography mission of Canada's National Quantum Strategy and the expertise the federal migration needs, complementing the Ko Lab's work on cryptographic hardware at the University of Saskatchewan — and is co-supervised with Dr. Sugata Gangopadhyay at IIT Roorkee, giving interns a direct research connection on both sides of the partnership.

What you'll bring:

Essential:

• Strong Python programming, including clean, tested code and batch experiments (NumPy, Git)
• Linear algebra: bases, Gram–Schmidt orthogonalization, determinants; comfort with modular arithmetic and polynomial rings
• Basic probability and statistics: distributions, estimating success rates, fitting curves
• Ability to read papers and documentation independently and keep a careful research log

Desirable (not required — will be taught):

• An introductory course in cryptography, number theory, or abstract algebra
• Exposure to lattices, LLL, or learning with errors (LWE)
• Experience with SageMath, Jupyter, the Linux command line, or a batch scheduler
• Some C/C++ (fplll, the library underneath fpylll, is written in C++)

Suitable students are undergraduates in computer science, mathematics, mathematics and computing, or electrical/computer engineering who enjoy both mathematics and experimental programming. Prior research experience is not required — the first two weeks include a structured reading and tutorial program on lattices, LLL/BKZ, and LWE, so a student who meets the essential requirements can contribute from week 3.

Department: Mathematics and Computer Science
Preferred background: Mathematics, computer science
Preferred start date: May 1, 2027 (flexible)

Professor Kharaghani's research specializes in combinatorial design theory, Hadamard matrices, graph theory, and related algebraic and computational methods — developing new constructions and classifications of structures such as symmetric designs, weighing matrices, conference matrices, orthogonal designs, Latin squares, and strongly regular graphs, with applications in coding theory, communications, cryptography, and experimental design.

Spectral graph theory and combinatorial design theory rely heavily on the spectral properties of classical adjacency and incidence matrices, but unsigned structures often obscure fine-grained algebraic constraints and extremal sub-configurations. Hao Huang's resolution of the Sensitivity Conjecture demonstrated the power of signed adjacency matrices, using an edge-signing of the hypercube to expose latent spectral gaps that had eluded unsigned methods for decades.

Building on this breakthrough, the project undertakes a systematic investigation into the algebraic and spectral theory of signed graphs, complex gain graphs, and signed incidence designs — enriching graphs, designs, and finite geometries by assigning weights from {−1, 1}, roots of unity, and compact or finite group elements to edges and incidences. By synthesizing techniques from algebraic combinatorics, finite geometries, matrix analysis, and representation theory, the project aims to establish a cohesive mathematical framework for signed combinatorial configurations.

What you'll bring:

A strong background in undergraduate mathematics, particularly linear algebra, discrete mathematics, and proof-based reasoning. Prior coursework in graph theory, combinatorics, abstract algebra, or matrix theory is desirable but not mandatory; you should be comfortable reading and writing mathematical proofs, working with matrices and eigenvalues, and learning new theoretical concepts independently.

Experience with mathematical software or programming (SageMath, GAP, Mathematica, Maple, Python, or MATLAB) is an asset, since computational experimentation may complement the theoretical work. Familiarity with finite groups, representation theory, finite geometries, or spectral graph theory is beneficial but not expected.

This project suits students interested in algebraic combinatorics, graph theory, design theory, finite geometry, or related areas of pure mathematics who bring strong analytical ability, intellectual curiosity, and persistence in tackling challenging problems.

Department: Mathematics / Chemical & Biological Engineering
Preferred background: Chemical engineering, mechanical engineering
Preferred start date: May 15, 2027 (flexible)

Professor Feng's research covers fluid dynamics, computation, cell and tissue mechanics, and transport phenomena.

Hydrogels are finding applications across emerging technologies, from organ-on-a-chip devices to the processing of cellulose nanofibrils. One of the most exciting applications is bioprinting with hydrogel-based bioink: hydrogels containing biological cells are extruded from an orifice and printed onto a substrate, where they solidify into a three-dimensional structure that scaffolds the cells as they proliferate and differentiate into functional artificial tissues or organs. Success depends on balancing two competing requirements for the bioink — low enough yield stress and viscosity to be easily printed from thin syringes ("printability"), and rapid enough re-solidification once deposited to hold the intended shape ("self-healing").

The rheology of the bioink, especially its yield stress, is central to the printing process. Gels undergo large-strain deformation while retaining most of the bonds in their solid network, with complete fluidization only occurring at much higher stress or strain. This internship uses a novel structure-based constitutive model to simulate the dynamics of hydrogel-based bioink during extrusion printing, capturing most of the experimentally observed yielding phenomena in hydrogels under shear. The intern will first adapt the shear-based model for general 3D flows, then use computational tools to study the bioprinting process in an axisymmetric 2D domain — aiming to understand how the ink yields during injection and resolidifies upon deposition, so that modelling and simulation can guide the design of desirable printed morphologies.

What you'll bring:

At least two years of undergraduate coursework in differential equations, fluid mechanics (or transport phenomena), and computational methods, plus experience with scientific computing — ideally from research projects touching these topics. Excellent English communication skills, both oral and written, are also expected.

Department: Mathematics
Preferred background: Mathematics, physics
Preferred start date: May 3, 2027 (flexible)

Professor Fetecau's research involves theoretical and numerical investigations of mathematical models for aggregation and swarming phenomena, with particular interest in applications to machine learning (transformers), population biology (swarming or flocking of animals), and robotics. Models in the literature range from particle-based (ODE) to kinetic and continuum (PDE) descriptions, most leading to nonlinear and nonlocal differential equations that are challenging to analyze and simulate.

Mathematical models of aggregation and swarming have drawn a surge of interest in recent years, largely to understand how self-collective behaviours emerge in groups of autonomous agents with no leader or external coordination — as in natural swarms (flocks of birds, schools of fish) and artificial mobile agents (robots). Despite extensive research on such models in Euclidean spaces, very little work has addressed aggregation models on arbitrary surfaces or manifolds, which are needed for engineering applications (e.g., robots constrained to a particular manifold by their environment) and for machine learning, where the self-attention mechanism in transformer architectures is modelled as an interacting particle system on a sphere.

The project investigates — theoretically and numerically — an aggregation model on surfaces and Riemannian manifolds: an aggregation-diffusion equation for the macroscopic population density, with interactions modelled via an interaction potential. Though expressed as a PDE, the model also has a discrete, individual-based ODE formulation enabling direct numerical experiments; the diffusion-free model has already been studied on the sphere, hyperbolic space, and the rotation group. The project aims for a better analytical and numerical understanding of the long-time behaviour and equilibrium solutions of this model, potentially including applications to transformers via the sphere case.

What you'll bring:

Solid mathematical training, particularly in analysis and differential geometry, together with familiarity with basic numerical analysis (numerical linear algebra and numerical integration of ODEs). Some coding experience in MATLAB is strongly desired.

Department: Chemistry
Preferred background: Mathematics, chemistry, physics
Preferred start date: May 31, 2027 (flexible)

Professor Bowles's research focuses on the theory and simulation of soft condensed matter — colloidal fluids, liquids, and glassy materials — whose unique thermodynamic and dynamic properties arise from the collective interactions of particles over mesoscopic length scales, using statistical mechanics, molecular dynamics, and Monte Carlo simulation. Current work focuses on confined fluids, where geometric restrictions on the length scale of mesoscopic interactions induce new and interesting phenomena.

Bulk liquid water exhibits thermodynamic and dynamic anomalies — a density maximum, increasing compressibility on cooling, increasing viscosity with pressure — that growing evidence links to a liquid-liquid critical point in the metastable region of the liquid-ice phase diagram. The transition between the two liquid states has in turn been related to topological changes in the hydrogen-bond network, where densification occurs through increasingly interpenetrating rings rather than through more neighbours.

Confining water to a quasi-one-dimensional channel, such as a carbon nanotube, changes the nature of this hydrogen-bond network topology. This project develops simple lattice models for network-forming liquids in confinement and studies their thermodynamic and topological properties, comparing simple volume-exclusion behaviour to open, directional, water-like bonding interactions. Taking advantage of the quasi-one-dimensional confinement, the project uses the transfer matrix method to access the partition function and equation of state as functions of temperature and pressure, as well as local lattice properties — and to generate equilibrium configurations for studying global bonding-network properties, including ring and knot statistics such as writhe and linking number, as functions of channel diameter.

What you'll bring:

A strong working knowledge of linear algebra, with some experience in computation. Background in thermodynamics (chemistry or physics) would be advantageous.

Department: Electrical and Computer Engineering / Biomedical Engineering
Preferred background: Electronics and communication engineering; electrical engineering; computer science; biomedical engineering; artificial intelligence; signal processing
Preferred start date: June 1, 2027 (flexible)

This lab's research focuses on hardware-efficient artificial intelligence, biomedical signal processing, and foundation models for resource-constrained computing platforms — self-supervised learning, EEG representation learning, Transformer architectures, model compression, quantization, and efficient inference on embedded systems and FPGAs — aimed at accurate, computationally efficient, deployable AI for edge-intelligent healthcare.

This project develops a compressed, multi-task EEG foundation model capable of performing four brain-signal analysis tasks: driver drowsiness detection, mental workload estimation, sleep-stage classification, and motor imagery recognition, exploring how pretrained EEG foundation models can be adapted for diverse biomedical and cognitive monitoring applications while meeting the computational and memory constraints of edge hardware.

The intern will first reproduce selected experiments using NeuroLM (ICLR 2025), leveraging publicly available implementations and pretrained weights, across four EEG datasets: DD-Database (driver drowsiness), STEW (mental workload), Sleep-EDF Expanded (sleep staging), and the PhysioNet EEG Motor Movement/Imagery dataset (imagined hand movement classification).

Following baseline reproduction, the intern will develop task-specific evaluation pipelines and investigate efficient transfer learning, exploring model compression techniques — structured pruning, low-bit quantization, and selective Transformer component reduction — to minimize computational complexity and memory requirements. Compressed models will be evaluated against their original counterparts on classification accuracy, macro-F1 score, parameter count, model size, memory utilization, and inference latency, with attention to which components can be simplified without substantially degrading performance.

Expected outcomes include reproducible benchmarking pipelines, a comparative analysis of compression strategies, and an optimized EEG foundation model suitable for subsequent FPGA or embedded hardware implementation. The project is co-supervised with Dr. Tharun Kumar Reddy Bollu at IIT Roorkee, giving interns a direct research connection on both sides of the partnership.

What you'll bring:

A senior-undergraduate background in electronics and communication engineering, electrical engineering, computer science, biomedical engineering, or a related discipline, with:

• Python programming, including familiarity with NumPy, Pandas, and SciPy
• Fundamentals of machine learning and deep learning, with PyTorch or TensorFlow
• Basic digital signal processing, including filtering and time-series analysis
• Systematic experiment design and model evaluation, plus basic Git and Linux skills

Experience with EEG or biomedical signal datasets, Transformer architectures and attention mechanisms, transfer learning, model pruning or quantization, knowledge distillation, or embedded/FPGA systems is a plus but not required — prior experience with EEG foundation models or FPGA development is not mandatory. The intern will receive guidance on EEG preprocessing, baseline reproduction, model adaptation, compression experiments, and performance benchmarking, and should be motivated to work independently, maintain reproducible research code, and communicate technical results clearly.