Explore Projects

<p>Under guidance from the mentors, the student will develop a suite of models for benchmarking Bayesian computation. That involves gathering data/models with a variety of inferential structure for coding and optimization in Stan. The resulting canonical models and posteriors will be submitted for inclusion in the posteriorDB database to serve as reference points against which new approaches to Bayesian computation can be compared.</p>

This project aims to improve the existing deep learning for computer vision ecosystem in Julia, especially by adding more models that users can use without much modification. Metalhead.jl is the package where this project will be working on consolidating all such computer vision models., The goal will be to add models that are both customisable and extensible so that they can be used for tasks such as image classification and semantic segmentation with ease.

<p>Estimations of a large family of statistical models can be casted as optimizing biconvex functions over given (bi)convex sets. Our goal in this project is to develop a generic computational framework for biconvex minimization via the alternating minimization algorithm, including sparse bigraphical models, sparse conditional graphical models and sparse matrix factorization. We describe the design and the implementation plan of our project, and propose a timeline for our development.</p>

<p>Currently, FluxML/model-zoo has very few examples demonstrating the usage of Flux. In this project, I propose to implement several Convolutional Neural Network Models for Computer Vision Problems. I shall be implementing around 14 Neural Network Architectures. I will train these models on public datasets to demonstrate their performance. In the process, I will contribute to Flux.jl the layers that are necessary to build those networks so that they can be easily reused.</p>

TARDIS uses a monte carlo method to propagate photon packets through a model of an exploded star. As these virtual packets move through the ejecta model they interact with the gas through absorption, emission, and scattering. TARDIS has the functionality to track the properties of the packets that are generated when running the Simulation. But currently, all this tracked data is stored as ASCII data. This project will build an interactive animated visualization tool so that scientists can see exactly how packets move through their models.

<p>The project aims implementation of GANs in the Machine Learning toolkit, TMVA of the ROOT framework would be immensely useful because of the advent, popularity and versatile nature of GANs. GANs can essentially be used for simulation and physical/mathematical modeling of patterns learned from training data substantially faster and more accurate than any other generative model. The model can be used for generating training data and finds many applications in high particle physics and astrophysical research realms.</p>

The OWASP DevSecOps Maturity Model is used to assess and present the devsecops maturity of an organization. It consists of an application and DevSecOps maturity model information. The application is used to present and assess the model itself. As it is aged, a new modern application with a frontend in angular is to be developed. Expected Results: The application should work in the browser only and is created with angular. The application should be able to assess multiple teams and present some statistics so that it is easy to get an overview of the maturity of all teams within an organization. The application has the current abilities and more

Projection predictive inference is a decision theoretic Bayesian approach that decouples model estimation from decision making. Given a reference model previously built including all variables present in the data, projection predictive inference projects its posterior onto a constrained space of a subset of variables. Variable selection is then performed by sequentially adding relevant variables until predictive performance is satisfactory. I propose to develop a Bambi-compliant module performing projection predictive model selection for GLMs.

The project aims to develop a comprehensive library of predefined force and torque models. These models are crucial for a wide range of scenarios in computational physics and engineering simulations. Traditionally, users have had to manually generate various forces and torques, a process that is often complex and error-prone. Our library will streamline this process by including commonly used models such as the Duffing Spring, Coulomb Friction, Hill-Type Muscle, and Aerodynamic Forces. This will greatly enhance the functionality and efficiency of SymPy.

The core of this project is a credit assessment engineer for Fineract, which provides client SDK and plugins to integrate into the platform. Phase 5 will be based on the work built before, which will more focus on updating the credit prediction methods used by the engine. There will be three main things to do:(1) Build a better ML and Stat classification model and delpoied them in the backend. (2) Select better features for the Rule-Based scoring model. (3) Setup a pipeline to auto update the data and train the model.

This project extends the gfpop R package to support time-dependent constraints. Instead of using a fixed constraint graph, the graph can change at each data point based on a user-defined rule function. This makes it possible to express models like LOPART and labeled up-down peak detection within a single framework. Deliverables include the R API extension, C++ implementation, oracle-validated LOPART model, a 6-rule up-down model with label support, and a user-facing vignette.

This project aims to improve probabilistic solar forecasting by replacing traditional quantile regression with a more expressive Gaussian Mixture Model (GMM) approach. Instead of predicting fixed quantiles, the model will learn the parameters of a GMM to represent the full predictive distribution, capturing complex and multimodal uncertainties common in solar irradiance data. Deliverables include a GMM-based forecasting model, performance comparisons with quantile baselines, robust uncertainty evaluation, and well-documented code for reproducibility and future extension.

<p>My project is divided into three parts:</p> <ul> <li>What is the Space of Spectral Sensitivity Functions for Digital Color Cameras</li> <li>Implementation of new colour.appearance model</li> <li>Color.models sub-package in "object"</li> </ul> <p>For the first two parts, my work will consist in studying scientific publications with the aim of recovering the mathematical formulas of the models to port them in python and for the third part my work consists in refactoring the python code in "object".</p>

Design a modular switched Ethernet model capable of simulating 1, 10, and 40 GbE links, with MAC-based auto-learning of I/O ports, support for pluggable scheduling algorithms (FCFS, WRR), while remaining extensible for future support of VLAN and Spanning Tree protocols, introduce enhanced EthernetNetDevice model with point-to-point–like NetDevice abstraction for switched Ethernet, an EthernetChannel capable of simulating both half-duplex and full-duplex modes of operation responsible for modeling transmission delay and errors.

<p>The aim of this project is to create a module to import geometric data acquired from terrestrial laser scanners. It will focus on developing a method to directly model real objects and topographies without entering their geometries manually. The information gathered by the laser scanner in the form of ‘Point Cloud Data’ will be used to create a full 3D model of the object. The data from the 3D model is required to be mapped/converted/translated onto the Finite-Difference Time-Domain (FDTD) grid upon which gprMax is based.</p>

<p>The cyclocopter module currently available in MBDyn models the inflow of a cycloidal rotor. It is however limited to 2D aerodynamics models; moreover, the implementation of the double multiple streamtube model is not yet stable. Calculate the inflow velocity vectors over the cyclocopter rotor (at each azimuth location) using the Double Multiple Stream Tube Method (Combination of Blade element theory and Momentum theory). Which will basically consider/capture the effect of induced velocity generated by upper half blades on the lower half blades.</p>

Currently, p5.js has a frustrating limitation for 3D artists: it flattens every imported model into a single shape that can only hold one texture. If you export a complex character from Blender or Maya with different materials for things like skin, clothing, and eyes, p5.js silently discards that information during the import process. This leaves creators with a flat, grey model instead of the detailed, textured work they actually designed. This project fixes the problem at its core by updating the library to recognize and render multi-material models automatically.

<p>This project follows Rahman and Vossmeyer (2019) as its motivating literature, and contributes to the three literatures by extending the various methodologies to a hierarchical Bayesian quantile regression model for binary longitudinal data (QBLD) and proposing a Markov chain Monte Carlo (MCMC) algorithm to estimate the model. The model handles both common (fixed) and individual-specific (random) parameters (commonly referred to as mixed effects in statistics). The algorithm implements a blocking procedure that is computationally efficient and the distributions involved allow for straightforward calculations of covariate effects.</p>

This project attempts to integrate DeepForest's object detection models into Large Language Models (LLMs) so that ecological images can be queried using natural language. Building on the pre-built models in DeepForest for tree, bird, and livestock detection, the system would extract structured annotations from images and convert them to an LLM-compatible format. Users can interact with an image through a queryable interface and receive proper responses based on object detection and spatial reasoning. This will involve prompt engineering, fine-tuning and experimentation with different LLMs.

Anomaly detection is the process of identifying data points, events, and observations that differ from a dataset's expected behavior, called (outliers). Researchers are interested in searching for rare events (new physics), a key feature of particle physics. When detecting these events using various methodologies, it is vital to have real-time and accurate models. Autoencoders and Deep Learning models play a core role in detecting anomalies with high efficiency. Therefore, this project aims to examine and evaluate different autoencoder models for detecting anomalies. And to compare their results with the Contrastive Learning.

<p>Ensemble methods combine predictions of base models in order to improve performance or generalizability over a single model. To this end, there are many approaches that either use averaging or voting over base model predictions. The main objective is to integrate ensemble support for the existing flow of the WSO2 Machine Learner. Implementation will include following tasks: Implement an ensemble method(s), to combine multiple algorithms (particularly Stacking, Boosting and Bagging) Create a UI to include the end to end flow of training the algorithm. Integrate it to the WSO2 Machine Learner.</p>

I propose to create a semantic multimodal search engine for collections of transcribed and aligned videos using state-of-the-art artificial intelligence models of different types, including NLP (Large Language Models) for text generation and capturing the semantics of transcriptions, as well as image description models to understand what is being seen in the video. Only focusing on the transcribed text does not help much, considering the context, what is being shown in the video will be helpful. The search engine will list down the most closest matches to the user query containing the metadata like link to the video, video-id, timestamp, text.

This proposal aims to address the challenge of I/O bottlenecks in High-Performance Computing (HPC) systems by developing a Graph Neural Network (GNN) model for accurate and interpretable I/O performance predictions. The approach includes creating a data pre-processing pipeline for I/O logs, constructing a GNN model tailored to the intricacies of HPC I/O, and implementing SHAP value analysis for insightful feature contribution understanding. Deliverables encompass the GNN model, integration with the AIIO framework, and a SHAP-based evaluation methodology to enhance diagnostic precision and accountability in I/O performance assessments.