Explore Projects

<p>Data visualization plays a crucial role in TVB's neuroinformatics platform, and a Structural Connectivity (connectome) is a core datatype, modelling full brain regions and their connections. It gives the main idea to manipulate the connectivity for better understanding in our research. Currently we have 2D visualizer which explain these connectivity. Here in this project we are going to implement 3D visualizer with refactoring the whole TVB’s front end from the UX design. Apart from this we are also going to stressed out on optimize these for extremely large data structure.</p>

<p>To guide behaviour, it has been proposed that neurons eventually learn to predict future states of sensory inputs. The project mentors have worked in this direction to get metrics on these predictions about how accurate those predictions are if the neuron used synaptic learning rules. The main contribution of this project would be to publish highly optimised library codes that can serve as evaluation benchmarks for predictive accuracy. We also believe that neurons can generate efficient encodings on these predictions. Through this project, estimates of the predictive information would also be obtained by neural models.</p>

Neurobagel is a federated data ecosystem that allows researchers and other data users to find and consume research data that has to remain at their original institute for data governance reasons. Currently, the researcher or the data user has to answer a number of queries to get the desired results and it often requires iteration. My aim would be to make this search process more user-friendly by adding an LLM style chatbot interface. This is to be done by utilizing large language models that will be able to interpret the user prompts and initiate the API calls accurately giving the user the desired results.

In some organisms like axolotl, the surface of the embryo is transparent and thus allowing us to see the embryogenetic events taking place before the neural tube closure. Therefore, acquiring images of the outside of early stage developing embryos could give us many insights. Hence, aim of this project is to build a computational tool that allows us to visualize 4D data derived from the surface of an Axolotl embryo using a special microscope called Digital Microsphere. That is, a computational tool which will display a sphere on which the images of the embryo from different angles will be mapped creating a 3D model.

The Virtual Brain (TVB) provides an open-source simulation framework for whole-brain network modeling. Unlike the traditional approach of using neurons at micro-scale level, TVB takes into account the brain regions and their interactivity. This proposal aims to enhance the TVB ecosystem by developing and improving interactive JupyterLab widgets — BCT metrics projection viewer, Connectivity_react widget and Unified Head widget. The new widgets proposed will extend TVB's analytical capabilities and improve the overall user experience for researchers studying large-scale brain connectivity patterns.

This proposal aims to improve the capability of markdown exporter package inorder to cover multicompartment model, and also to support generation of p.d.f, h.t.m.l , latex file types not only this but also to follow a standard template generation using jinja2 templates , Inorder to streamline the process for user to select the output according to their requirements based on their template selection. apart from this handling the python environment inorder to extend support for morphologies , moreover allowing users to use their own templates from their file system, which enables them to make the best use of brian2tools package.

<p>Image registration is the process of finding a transformation that aligns one image to another. DIPY currently supports several numerical optimization-based techniques for image registration. Even though these methods perform well, they are limited by their slow registration speeds. The goal of this project is to develop deep-learning-based methods that can achieve image registration in one shot, resulting in much faster registration speeds. In this project, I propose to develop several deep neural networks for affine and deformable MRI registration. Additionally, I also plan to implement thin-plate splines.</p>

SustainHub is an AI-driven simulation platform designed to study and enhance the sustainability of open-source communities. It models contributors, tasks, and decision-making using reinforcement learning to mimic real-world collaboration dynamics. By integrating methods like Multi-Armed Bandits for task allocation and SARSA for adaptive learning, SustainHub provides insights into workload balance, contributor retention, and project health. The system introduces quantitative metrics such as the Harmony Index to measure long-term sustainability. Ultimately, SustainHub helps researchers and organizations design strategies for thriving, resilient open-source ecosystems.

<p>Nighres is an open-source Python package that enables high-resolution neuroimaging data to be easily and efficiently processed. Diffusion MRI is an imaging modality that is sensitive to random displacements of water molecules in brain tissue, yielding valuable information about the size, shape, and orientation of neurons. The purpose of this project is to help the neuroscience community by efficiently implementing an automated white matter parcellation algorithm in Nighres, which can be easily used without technical expertise, and by improving the algorithm to work with more realistic diffusion models than diffusion tensors. A comprehensive documentation and a tutorial will be written for the method to facilitate its use.</p>

The Human Neocortical Neurosolver (HNN) is a computational tool that enables scientists to study brain responses at the cellular and circuit levels. However, the current version of HNN-core has a limitation in that it cannot simulate large batches of simulations. This hinders the efficient optimization of parameters, which is crucial for accurate modeling. To address this issue, I am working on improving HNN-core by adding batch simulation functionality. I am also developing a user-guide on advanced optimization techniques, specifically focusing on simulation-based inference (SBI), a deep learning-based Bayesian inference method. These enhancements will significantly contribute to the HNN's codebase, enabling researchers to extract crucial insights for developing theories about human brain response origins.

GeNN is a C++ library used for simulating Spiking Neural Networks through GPU-based code generation (CUDA/HIP). Currently, its accessibility is limited to users with a dedicated GPU system. This project aims to address this limitation by adding an ISPC (Intel’s Implicit SPMD Program Compiler) code generation backend model to the GeNN library to enable efficient SNN simulations on SIMD (Single Instruction, Multiple Data) capable CPUs . My project will expand GeNN’s accessibility to those users who do not have a dedicated GPU system while delivering high performance through CPU SIMD parallelism. I will work on backend architecture design, SIMD optimization, build system integration, and benchmarking against existing GPU implementations .

SciCommons is an open source platform for researchers to interact, review, and discuss scientific papers. My project focuses on making the platform a more powerful tool for daily research. I will build a browser extension that lets users import articles instantly from sites like arXiv, PubMed, and Nature. I will also build an AI summarization feature using locally hosted Ollama models to help researchers process papers faster. To keep the site fully accessible for everyone, I will expand our automated testing coverage using PlayWright and Axe-core. Finally, I will integrate real-time social feeds on home page and a new split-pane preprint viewer to improve how we discover and read research papers.

<p>Neuronal and multiscale simulations are computationally demanding, they require large numbers of very similar calculations. One of the core problems in this domain is to rapidly perform detailed single-neuron calculations. These are typically the bottleneck in detailed multiscale and network models. The central computation is solution of a large, almost tridiagonal matrix representing compartments in a neuron. Individual entries in this matrix require an inner loop to compute current contributions to the compartment. It is a particularly interesting problem to optimize GPU computations for these neuron calculations, since there is a tradeoff between memory transfers and speed of individual GPU cores. Optimizing the GPU code will increase the speedup, when the no of computations are very large, then even a small amount of speedup is very beneficial as it saves a lot of time, so optimizing the GPU code is important.</p>

This project accelerates the convolution core of the Active Segmentation Plugin (ASP) in ImageJ by offloading its most compute-intensive pixel operations to parallel hardware using TornadoVM. The goal is to achieve significant speedups without altering existing functionality or behavior. The approach focuses on identifying CPU bottlenecks, converting them into accelerator-friendly kernels, and executing them through a reusable TornadoVM pipeline. A strict correctness-first methodology ensures all accelerated outputs match the CPU reference in terms of boundary handling and numerical accuracy before any performance gains are considered. By integrating a robust, reusable parallel execution layer with safe CPU fallback, this work transforms ASP’s performance while preserving its reliability—laying the foundation for scalable acceleration across multiple filters and larger image-processing workflows.

<p>PyNN is a project written in python that aims at interfacing a handful of neural network simulators. A unified high level interface means that a model can be configured easily and only once and then it can be deployed on all supported simulators. Different simulators provide diverse features and also might use various numerical approaches and approximations leading to possibly different results. Having an option to compare data from several simulators without extra work is a great feature PyNN offers.</p> <p>GeNN is an efficient neural network simulator written in C++. The characteristic feature of this simulator is the ability to run simulation on GPU thus greatly decreasing required computational time.</p> <p>Interfacing GeNN from PyNN would be a valuable acquisition for both projects. PyNN will have another, faster simulator at hand and GeNN will get a python interface. Python is much more widespread among the neuroscientists. Having such an interface will help to make GeNN more popular and hopefully will reduce the time scientists spent on their simulations.</p>

The Human Neocortical Neurosolver (HNN) is open-source, computational neural modeling software that allows us to examine the cellular- and circuit-level basis of brain responses. HNN requires the hand-tuning of a large set of parameters until a close fit between simulated and recorded data is attained. This hand-tuning can take a substantial amount of effort thus it is in the user’s best interest to automate the process so that parameters can be optimized in a time efficient manner. The goal of this project is to develop optimization functions that will perform a wide search over the parameter space to arrive at faithful simulations. HNN is currently being used to develop or test hypotheses about underlying circuitry that gives rise to cognitive processes of interest. Developing a robust algorithm for parameter optimization has the potential to illuminate avenues for the diagnosis and treatment of multiple brain disorders and diseases, cognitive impairment, and psychiatric disorders. Finally, contribution to HNN’s codebase will aid researchers who use the tool in yielding important constraints to the development of theories about the origins of human brain responses.

This project will involve refactoring the Human Neocortical Neurosolver (HNN-core) codebase to enable flexible development of neural circuits with customizable cell types. Currently, HNN-core contains hard-coded assumptions about cell types, locations, and connectivity that limit researchers' ability to implement networks with different architectures. I will restructure key modules to replace these hard-coded constraints with dynamic definitions, allowing users to flexibly select and connect cell types with different morphologies and electrophysiological properties. The solution involves refactoring cells_default.py for dynamic cell creation, modifying network.py to handle arbitrary cell types, and updating related components to support custom cell types. The proposed plan includes dynamic cell type implementation, updated network connectivity handling, a demonstration model with custom cell types, comprehensive documentation, and unit tests ensuring backward compatibility. This work will significantly enhance HNN's capabilities for hypothesis testing in EEG/MEG research while maintaining its user-friendly approach for neuroscientists.

Last year, work on GestureCap focused on developing an Arduino-based latency measurement system and optimizing the pipeline for low-latency performance. Additionally, OSC-based communication between Python and PureData was implemented to enable real-time audio processing. Building on this foundation, this project will introduce AI-driven gesture recognition using lightweight, GPU-accelerated models to improve accuracy and responsiveness. The gesture-music mapping will be expanded to support both trigger-based events (e.g., drum beats) and continuous controls (e.g., pitch, reverb, and panning), allowing for more expressive musical interactions. The project will also integrate GestureCap with MaxMSP and Ableton Live, making it more accessible to musicians and producers. Further improvements will focus on latency reduction, real-time inference optimizations, and testing OpenPose-based feature extraction as a potential enhancement. If time permits, a visualizer for live gesture-music interaction will also be developed. These advancements will refine GestureCap’s capabilities, ensuring a more responsive and expressive experience for users.

<p>ImageJ is an OpenSource image processing tool written in Java which has been extremely helpful in the analysis of scientific images especially medical and microscopic. ImageJ package consists of a plugin for segmenting all the cells in an embryo of a C.Elegan from raw SPIM images. This plugin which has been built as a part of GSoC’17 comprises of standard image processing techniques to segment the cells.</p> <p>While attempting to segment cells from High-Resolution time-lapse movies of Embryogenesis, the used unsupervised algorithms like Intensity thresholding, Watershed Transform, Active Contours, etc, which depend on extracted local or global features, fail to be resistant against uneven illumination, optical noises, complex cellular shapes and more of such distractions. Thus, we are still in the pursuit of having a robust method for accurately segmenting microscopy images.</p> <p>The top priorities of this proposal are: Extend the plugin’s capability to more accurately segment the cells by adding Semi-Supervised and Unsupervised methods. Focus on extending capabilities of tools for developmental DataScience with the help of datasets or develop a cell-tracking system for Bright-field movies.</p>

HarmonyHub is an open-source, modular web application designed to revolutionize music education by integrating modern technologies and pedagogical principles. Built using Ionic and Angular, it provides educators with reusable components and services to create adaptive and personalized learning experiences. Initially focused on trumpet players, the platform is now being expanded to support other variable-pitch instruments like wind (e.g., Flute, Clarinet) and bowed string instruments (e.g., Violin, Viola). By blending traditional music instruction with cutting-edge digital tools, HarmonyHub enhances practice strategies, making learning more engaging and effective. Mastering wind and bowed string instruments is challenging, requiring structured practice and targeted support. HarmonyHub addresses this need by bridging the gap between conventional teaching methods and modern technological advancements, ensuring students receive interactive and personalized guidance. In an era where digital transformation is redefining education, integrating Information and Communication Technology (ICT) into music learning enhances skill development, creativity, and motivation. By refining its interface and expanding its instrument coverage, HarmonyHub paves the way for a more inclusive, accessible, and innovative approach to music education.

Event-based vision is a subfield of computer vision that deals with data from event-based cameras. Event cameras, also known as neuromorphic cameras, are bio-inspired imaging sensors that work differently to traditional cameras in that they measure pixel-wise brightness changes asynchronously instead of capturing images at a fixed rate. Since the way event cameras capture data is fundamentally different to traditional cameras, novel methods are required to process the output of these sensors. In addition to dealing with ways for capturing data with event cameras, event-based vision encompasses techniques to process the captured data - events - as well, including learning-based techniques and models, spiking neural networks (SNNs) being an example. This project aims to create benchmark datasets for object recognition tasks with event-based cameras. Using machine learning solutions for such tasks requires a sufficiently large and varied collection of data. The primary goal of this project is to develop Python utilities for augmenting event camera recordings of objects captured in an academic setting in various ways to create benchmark datasets. A secondary goal of this project is to test the performance of spiking neural networks for object recognition on the created datasets.

This project aims to add brain imaging data structure (BIDS) provenance to a workflow engine written in Python called ‘Pydra’. The brain imaging data structure format is a widely-adopted framework for disseminating clean datasets for use in the reproducibility of neuroimaging data analyses (https://bids.neuroimaging.io/). This project would aid in this pursuit by establishing built-in support for the BIDS format within the Pydra software package. Pydra is a library that allows complete control of the command line through Python (e.g., within a Jupyter Notebook). The majority of brain imaging data is analyzed using many commands that operate in a serial manner for cleaning, aligning, and applying statistical models to the data often in batches. Rather than relying on shell scripting and manual documentation of software versions, commands used, and outputs that are generated, Pydra can do so automatically. Pydra not only allows commands to be run through Python that would normally be run within the command line, but also unlocks the ability to interface with any CLI binaries a user might have installed as third party tools. Beyond this, Pydra allows the development of ‘workflows’ or stepwise processing pipelines with traceable software and command provenance, allowing an individual to essentially run an exact analysis in a “frozen” environment that the original publisher/researcher/scientist used themselves.

MRI quality control (QC) is an essential but highly manual and repetitive process in neuroimaging research. Currently, reviewers must navigate scattered directories and mentally integrate diverse outputs, such as 3D volumes, SVG montages, and isolated metrics, to make a reliable assessment. This "hidden labor" takes significant time away from actual research, introduces variability across different raters, and results in decisions that are poorly documented and difficult to audit. This project aims to transform QC-Studio from a working prototype into a robust, Nipoppy-integrated application. The core of the solution focuses on replacing hardcoded file paths with a scalable, config-driven architecture utilizing Nipoppy's DatasetLayout API. Additionally, the project will inject essential quantitative context into the review process by building an Image Quality Metric (IQM) distribution panel. Finally, it will establish a structured "Evidence Bundle" to experimentally evaluate whether a lightweight LLM-based assistant can improve review efficiency by summarizing quantitative data and flagging anomalies. Key Deliverables: 1) Config-Driven Architecture: A unified Streamlit application featuring dynamic subject loading, unified pagination, and manifest.tsv integration across multiple processing pipelines without hardcoded paths. 2) Quantitative Evidence Layer: An interactive IQM panel that visualizes subject-level metrics against local datasets and crowdsourced reference distributions (MRIQC Web API), complete with rule-based outlier detection. 3) Experimental AI Guidance Prototype: A foundational integration displaying pre-generated, LLM-authored summary reports within the dashboard, accompanied by an evaluation report assessing factual accuracy and workflow impact.