Explore Projects

This document outlines a proposal to support Volcano for gang-scheduling in Kubeflow Trainer, so as to provide users with more AI-specific scheduling capacities like priority scheduling and queue resource management. This will allow for better control over resources, with features like queue management and job prioritization. Thanks to the Kubeflow Trainer Pipeline Framework, we can seamlessly integrate Volcano into Kubeflow Trainer as a runtime plugin. The deliverables include: 1. The integration of the Volcano plugin into the Trainer, allowing for Gang-Scheduling and resource management. 2. Documentation updates, including usage examples and design details. 3. Testing the plugin's functionality and performance under various environments.

Our "Support Zvk in T1" project aims to upgrade the T1 processor by adding Zvk, a cryptographic feature from the RISC-V architecture, enhancing its data processing and security capabilities. We'll start with researching how Zvk can fit into the T1's design, then move on to detailed planning and design. The heart of the project is the RTL design and implementation, where we'll turn our plans into a functional model. Testing ensures everything works perfectly, leading to the development of a demo application to showcase the new features. Documentation throughout will provide a blueprint for future advancements. This enhancement will not only improve the T1 processor's performance but also its ability to securely handle data, setting a new standard in the field.

CellMinerCDB provides a web-based resource for integrating multiple forms of pharmacological and genomic analyses, and unifying the richest cancer cell line datasets (the NCI-60, Sanger/MGH GDSC, and Broad CCLE/CTRP).CellMinerCDB enables data queries for genomics analyses and exploration of pharmacogenomic determinants of response. It leverages overlaps of cell lines and drugs across databases to examine several biological questions. CellMinerCDB is built using R Shiny which provides a modular framework. This proposal is aiming towards adding an additional module in CellMinerCDB for Network Analysis that will provide an easier visualization of averaged genetic profiling data on top of a curated set of pathways.

MapMint is a web-based Geographic Information System (GIS), which is designed to facilitate deployment of Spatial Data Infrastructure (SDI). In an SDI, geographic data, metadata, tools, and the users are connected in an interactive manner in a framework so as to use the spatial information in an efficient and flexible way. MapMint combines various different software in a complete and coherent web mapping platform, thus helping users in building their own maps and web-applications. These web-services are built on top of the ZOO-Project. Visualizing geospatial data in front end is a challenging task. For this, adding Cesium support within MapMint is a great task of building interactive web interface. My deliverables would be, Complete integration of Cesium Library within MapMint.

<p>Unlike traditional inverse identification tools that rely on gradient and gradient-free methods, simulation-based inference has been established as the powerful alternative approach that yields twofold improvement over such methods. Firstly, it does not only result in a single set of optimal parameters, rather simulation-based inference acts as if the actual statistical inference is performed and provides an estimate of the full posterior distribution over parameters. Secondly, it exploits prior system knowledge sparsely, using only the most important features to identify mechanistic models which are consistent with the measured data. The aim of the project is to support the simulation-based inference in the brian2modelfitting toolbox by linking it to the <code>sbi</code>, <code>PyTorch</code> powered library for simulation-based inference, development of which is coordinated at the Macke lab.</p>

ConformalPrediction.jl is a package for Predictive Uncertainty Quantification through Conformal Prediction for Machine Learning models trained in MLJ. This project aims to enhance ConformalPrediction.jl by adding support for the Conformal(ized) Bayes technique. The conformalized Bayes technique integrates conformal prediction with Bayesian inference to create prediction intervals that have the desired frequentist coverage properties while still being informed by the Bayesian posterior distribution. The integration will be carried out while maintaining consistency with the existing codebase, ensuring seamless incorporation of new functionalities. In addition to this primary objective, new methods for assessing the quality of probabilistic estimates and new recalibration method will be integrated into LaplaceRedux.jl, which is a library written in pure Julia that can be used for effortless Bayesian Deep Learning through Laplace Approximation (LA).

<p>For running its simulations TARDIS needs atomic data, such as atomic masses, ionization energies, levels and transitions. This data is available from a number of known sources. In order for TARDIS to be able to use the data, it should be downloaded, stored in a database, and converted to the HDF5 format. This task is currently done by the <code>tardisatomic</code> package; however, this package has some underlying issues. It has no clear workflow and the existing database structure is difficult to modify and maintain. These problems complicate adding and processing data from new sources.</p> <p>If my project is accepted, I will implement a new package named <code>carsus</code> (the name suggested by the collaboration) that will be deprived of the issues inherent in <code>tardisatomic</code>. The new package will have clear workflow, sound database structure and flexible input/output modules.</p>

<p>The current project aims to implement the option to carry out the numerical integration part in the Brian simulator with integrators defined in external libraries, starting with the odeiv2 module that comes with the GNU Scientific Library (GSL). In addition to coming with an extensive list of different integration methods, the GSL integrator comes with an adaptable timestep. The latter functionality will have major benefits for the speed with which large simulations in Brian can run. This is because it allows the use of larger timesteps for the overhead loops in Python, without losing (likely even improving) the accuracy of the numerical integration at points where small timesteps are necessary. I have personally used Brian, and can tell from experience that this can be very useful at times - especially when working with Hodgkin-Huxley-type equations. I think this speed-up could be an important factor for people to choose to use Brian for their simulations.</p>

This project aims to enhance the user experience for both challenge hosts and participants by developing an intelligent, RAG (Retrieval Augmented Generation) based chatbot. The chatbot will efficiently address queries related to challenge hosting, guidelines, troubleshooting, and FAQs. By integrating state-of-the-art NLP techniques with robust retrieval mechanisms, the solution will ensure prompt, accurate, and context-aware responses that reduce support overhead and streamline communication. Using the RAG approach, the chatbot will retrieve relevant information from challenge documentation and combine it with generative models to create coherent and helpful answers. This will empower hosts to manage challenges more effectively and assist participants in resolving queries, ultimately contributing to a smoother and more interactive challenge experience. Deliverables include : RAG Framework Implementation Chat UI Comprehensive Documentation Testing, Feedback, and Continuous Improvement

This project aims to integrate Logarithmic Number Systems (LNS) into a deep-learning framework, offering an alternative to traditional floating-point arithmetic. LNS simplifies multiplication and division, potentially reducing power consumption and improving efficiency in applications like deep learning, where approximate results are acceptable. However, current deep-learning frameworks, such as PyTorch and TensorFlow, are hardcoded to use floating-point (FP) operations, limiting the adoption of LNS. We seek to bridge that gap by developing a system that supports LNS arithmetic while maintaining compatibility with PyTorch. By creating custom layers, optimizers, and benchmarking tools, the project will enable the simulation of LNS arithmetic in FP hardware for neural network training and inference. The expected outcome is a robust, open-source solution to make research and exploration of LNS in deep-learning applications easier, paving the way for future hardware implementations.

Rebuilderd is a verification system for binary packages. It repeates the build process of a package in an identical environment and verifies that the package is identical. It currently generates in-toto link attestations when a package is successfully rebuilt. As part of this task, rebuilderd must be updated to generate in-toto SLSA provenance. To enable this feature, in-toto-rs must be extended to support the provenance specification as well. We plan to add a additional json generator, work together with existing generator by extract a abstract interface. By adding a argument, software will dynamic dispatch to right generator. By this way, in-toto-rs can be compatible with newer generators introduced by the future standards. After this, we will update the version in-toto-rs relied by rebuilderd, together with enough testcases to ensure the correctness of the logical. If there's time left, we will add a "Verifylib"(or part of it) into `in-toto-rs to make up for the current functional defects. This verifylib should be compatible with the attestation of both old and new version.

This project aims to address the challenge of creating a seamless and efficient integration between Gemini and Vertex AI in Firebase for Android and iOS platforms using Kotlin Multiplatform (KMP). The proposed solution is to develop an open-source KMP library that will support Gemini with Vertex AI in Firebase, focusing on best practices for creating KMP libraries and ensuring robust production implementation across both Android and iOS platforms. The deliverables for this project will include: 1. Kotlin Multiplatform Library: This library will provide seamless integration between Gemini and Vertex AI in Firebase, supporting Android and iOS. 2. Sample Code and Documentation: Along with the library, sample code and thorough documentation will be provided to assist developers in utilizing the library effectively for their own projects. In practice, developers can integrate this library into their Kotlin Multiplatform applications. Our library will offer a seamless process of integrating Vertex AI into Firebase apps for both iOS and Android.

LitmusChaos is an open source Chaos Engineering platform that enables teams to identify weaknesses & potential outages in infrastructures by inducing chaos tests in a controlled way. This project idea involves upgrading the Chaos Operator and Chaos Exporter repositories by updating their dependencies, addressing security vulnerabilities, and adding new functionality. Specifically, the project aims to upgrade the operator-sdk and Prometheus exporter versions, add new Prometheus metrics to the Chaos Exporter, and fix security vulnerabilities pointed out by trivy and golangci-lint. Furthermore, the project seeks to add unit test cases to both repositories to ensure that their functionality is robust and reliable. Overall, this project aims to improve the stability, security, and functionality of the Chaos Operator and Chaos Exporter repositories, making them better suited for use in production environments. - Expected outcome: The outcome of this project will be improved stability, security, and functionality of the Chaos Operator and Chaos Exporter modules through the upgrade of dependencies, addition of new metrics, and implementation of unit tests.

<p>This project aims at designing and implementing new features in Distro Tracker [<a href="http://tracker.debian.org/" target="_blank">http://tracker.debian.org/</a>] to better support Debian teams to track the health of their packages and to prioritize their work efforts. For this purpose, I will migrate the most important team-related features from PET [<a href="https://pet.debian.net" target="_blank">https://pet.debian.net</a>] to Distro Tracker, leveraging and improving Distro Tracker current code base regarding teams. Thus, as a final result of SoC, I expect to incorporate to Distro Tracker a set of useful data to help teams to see the health of multiple packages and better prioritize their efforts where it is most needed. It is worthing noticing that Distro Tracker is a general purpose service that is also used by Kali community [<a href="https://pkg.kali.org/" target="_blank">https://pkg.kali.org/</a>]. Thus, they also will be able to take advantage of the proposed improvements.</p>

<p>An asymptotic expression typically contains exact terms and O-terms, for example n3 + 2n2 + O(n). The basic framework for this asymptotic ring is already implemented. The main aim of this summer of code project is to extend its functionality up to the point where asymptotic expressions with explicit constants are fully supported. In his book, “Asymptotic Methods in Analysis” on page 5, deBruijn calls it an “L-term”, but we will call it “B-term”. B_{20}(3z) is a B-term, which stands for an expression which is bounded in absolute value by 3|z| for |z| ≥ 20. Sometimes, in research, it is important to know from which values of the variable on the asymptotic expression has enough precision to decide wether it is larger than a given bound. This cannot be achieved by O-terms (because the unknown implicit constant might be very large). For example, the average number of comparisons of classical quicksort is 2n log n + O(n). When comparing with recent variants, such as dual pivot quicksort with average number of comparisons 1.8n log n + O(n), it is interesting to know how large the O-terms are in order to know when dual-pivot quicksort will be more efficient than classical quicksort.</p>

pocketpy is a lightweight Python interpreter primarily designed for game scripting, boasting a vast feature set encompassing various Python modules. It's competitive performance and elegant syntax have made it a user choice for scientific computations. To enhance its capabilities, this project focuses on adding the core feature set from the NumPy library to pocketpy. There are two primary objectives: 1. Implementing core classes from NumPy: This subtask aims to implement key classes such as ndarray and random, along with their respective methods and functionalities. The code will be written in C++ for performance benefits. Existing C++ libraries like NumCpp (or Xtensor), which offer NumPy-like functionalities, will be leveraged for the implementation. 2. Exposing the C++ implementation to Python through bindings: To achieve this goal, we will use a library named pybind11, a popular header only library used to generate bindings for C++ classes automatically. This step will allow users to work on a pythonic frontend and access the NumPy functionalities implemented in C++. By the end of this summer, pocketpy seeks to offer robust array operations and random number generation capabilities, positioning it as a versatile tool for various computational tasks beyond game scripting.

<p>The Essential Care for Every Baby (ECEB) is an educational and training program, developed by the American Academy of Pediatrics, provides knowledge, skills, and competencies to nurses and doctors in low/middle-income settings so that they can provide life-saving care to newborns from birth through 24 hours postnatal.</p> <p>The aim of this project is to develop an Android mobile application, the ECEB Digital Action Plan, that is purposely built to provide clinical decision-support for nurses and doctors delivering essential newborn care interventions during the first day of life.</p> <p>This project aims to develop an ECEB Android Application with the following features-</p> <ol> <li>Time-stamping births to generate an essential newborn care clock for babies. </li> <li>Ability to track multiple babies at a time. </li> <li>Automated classification of babies health status, based on results from observations and assessments.</li> <li>Automated advice on the management of babies, based on their classification. </li> <li>Reminders/alerts to provide recommended essential newborn care interventions. </li> <li>If particular interventions are not delivered on time then the capture of information regarding why this essential newborn care was late or missed.</li> </ol>

<p><strong>MapMint4ME</strong> is an android application which allows its users to record alphanumeric data, photo and GPS locations. The data can be recorded even when there is no internet connectivity. When the user returns to a location with the internet connectivity, the data is uploaded to the database. MapMint4ME is very close to <a href="http://mapmint.github.io/userguide-fr/index.html" target="_blank">MapMint</a> web software which is built on the <strong>ZOO-Project</strong>. The <a href="http://zoo-project.org" target="_blank">ZOO-Project</a> is an SDI manager providing the capability to built map and web applications.</p> <p>In this project, I plan to <strong>add audio and video data recording</strong> facility to MapMint4ME. The main idea is to <strong>include SOS input support to MapMint4ME</strong> which will <strong>enable recording of sensor data directly within the app</strong>. The sensor data will be obtained from the sensors embedded on the Android platform. This project will help lots of GIS users, geologists, geographers <em>et cetera</em> who wish to retrieve data from remote areas (where it is not possible to reach) by accessing sensor data directly through the app. Certainly, more functions with extended capabilities could be added in future. I believe that by following a strict timeline as mentioned in the proposal, I will be able to finish the project successfully.</p>

ROOT's TMVA( Toolkit for Multi-Variate Analysis ) has SOFIE (System for Optimized Fast Inference code Emit) which offers a parser capable of converting ML models trained in Keras, PyTorch, or ONNX format into its own Intermediate Representation, and generates C++ functions that can be easily invoked for fast inference of trained neural networks. It is currently implemented for CPU inference along with a SYCL implementation. This project aims to explore different GPU stacks (such as CUDA, ROCm, ALPAKA) and implement GPU-based inference functionalities in SOFIE. Although there exist CPU implementations, working with HEP applications, need for GPU inferences become important. We will be writing optimized GPU Kernels for various operators such as RELU, GEMM, CONV2D, etc. that SOFIE already has for CPU implementation. Testing and experimenting which tech stack (such as CUDA, ROCm, ALPAKA) will be best aligned with SOFIE is our first goal. After getting to know about the pros and cons we can decide which one aligns with our goal. Our next aim would be to integrate all operators for GPU implementation in a incremental fashion ensuring each operator gets enough testing for edge cases. Once we have done basic implementation, we move on to optimizing thread-block config by using tools like Nsight for maximum performance. Once we have implemented our operators for the GPU stack, there is a need for performance benchmarking against popular frameworks like Pytorch/Tensorflow. Last part of the project would be extensive testing and benchmarking memory usage and execution time. I would end the project with proper documentation on how exactly to implement and scopes for further improvements.

<p>Meshery Models are declarative representations of infrastructure, applications, and their relationships - the canonical artifacts through which Meshery understands and manages cloud native systems. Today, Meshery lacks a standardized, portable distribution mechanism for these models. OCI registries (Docker Hub, AWS ECR, GitHub GHCR, and others) have emerged as the universal artifact store for the cloud native ecosystem, and [ORAS](https://oras.land) (OCI Registry As Storage) provides the Go-native tooling to push and pull arbitrary artifacts to any OCI-compliant registry. This internship implements end-to-end OCI registry support for Meshery Models - from new Connection and Credential types for major registries, to ORAS-powered push/pull logic in the Meshery server, to a redesigned Registry page in the Meshery UI that gives users full visibility and control over their model artifacts across registries.</p><p><br></p><p>Recommended Skills: Golang, REST API development, React. Familiarity with OCI image specifications, container registries, or ORAS is a plus. Experience with Meshery or other CNCF projects is welcomed but not required.</p><p><br></p><p>Responsibilities:</p><p>&nbsp;- Design and implement Connection and Credential types for Docker Hub, AWS ECR, GitHub GHCR, and additional OCI-compliant registries within Meshery's existing connection framework.</p><p>&nbsp;- Implement Golang server-side logic using the ORAS SDK to push and pull Meshery Models (and their component schemas, relationships, and policies) to and from any OCI-compliant registry.</p><p>&nbsp;- Define the OCI artifact media types, manifest structure, and layer conventions used to package Meshery Models for registry storage.</p><p>&nbsp;- Enhance or rewrite the Registry page in Meshery UI to surface connected registries, browsable model artifacts, push/pull controls, and credential management.</p><p>&nbsp;- Write integration tests covering push, pull, and round-trip fidelity of Meshery Models across at least two registry backends.</p><p>&nbsp;- Document the new registry integration, artifact format, and UI workflows in Meshery's official documentation.</p><p><br></p><p>Expected Outcome:</p><p>&nbsp;- Meshery users can connect to Docker Hub, AWS ECR, GHCR, and other OCI registries using managed credentials and push or pull Meshery Models directly from the Meshery UI and `mesheryctl`.</p><p>&nbsp;- A well-defined OCI artifact convention for Meshery Models, documented and suitable for adoption by the broader Meshery ecosystem.</p><p>&nbsp;- A redesigned Registry UI page providing a unified, registry-agnostic interface for model artifact management.</p><p><br></p>

This project focuses on completing and refining KDE Print Manager’s transition to CUPS 3.x, ensuring full compatibility with the newer printing architecture while still supporting CUPS 2.x. Although core functionality already works, several features—particularly those related to modern, driverless printing—remain incomplete. The shift in CUPS 3.x toward IPP-based print destinations instead of permanent queues requires KDE Print Manager to better handle discovery, grouping, and display of printers such as network devices, IPP-over-USB printers, Printer Applications, and shared remote queues. The interface should clearly reflect the temporary, on-demand nature of these queues. A major aspect of the work involves improving support for Printer Applications, which are replacing traditional PPD-based drivers. The system should prioritize these modern solutions in workflows like “Add Printer,” while still allowing legacy configurations when necessary. Additionally, the user interface needs refinement to clearly distinguish between classic queues, driverless printers, and Printer Applications, and to provide easy access to external configuration tools like web interfaces. Finally, the project emphasizes strengthening test coverage, especially for newer CUPS 3.x features such as IPP destination handling and Printer Applications. By integrating comprehensive unit tests into the CI pipeline, the goal is to ensure long-term stability and reliability as the Linux printing ecosystem continues to evolve.

<p>The goal of the project would be to add support for PSIMI-TAB format to the admin tools of the DV-IMPACT database and allow the upload and download of PPI network data in the same format. Currently, the data upload can only be performed by the database admin. The objective is to create registration and user profile system that allows users to register and upload their own data independently without the need of the database admin to be involved in the upload process.</p> <p>The summarized goals would be:</p> <ol> <li>Create registration and user profile system that allows users to register and upload their own data independently.</li> <li>Add support for PSIMI-TAB format to the admin tools of the DV-IMPACT database.</li> <li>Add download PPI networks in PSIMI-TAB format to the frontend of DV-IMPACT database.</li> </ol> <p>The Project fulfills all these issues by:</p> <ol> <li>Creating a User Profile system type forum.</li> <li>Allow users to register, upload and download domain-peptide PPI data in the specified format.</li> <li>Modify the DV-IMPACT database present in the server to handle the upload to the admin tools in PSIMI-TAB format.</li> </ol> <p>Link to the evaluation task: <a href="http://dvimpact.herokuapp.com/" target="_blank">http://dvimpact.herokuapp.com/</a></p> <p>Github Repo: <a href="https://github.com/jajodiaraghav/DV-IMPACT" target="_blank">https://github.com/jajodiaraghav/DV-IMPACT</a></p>

The goal of this project is to enhance the Katib Experiment APIs to support various parameter distributions such as uniform, log-uniform, and qlog-uniform. This improvement aims to align Katib more closely with other hyperparameter tuning frameworks like Hyperopt, which offer a broader range of parameter distributions. Currently, Katib is limited to supporting only uniform distribution for integer, float, and categorical hyperparameters. By introducing additional distributions, Katib will become more flexible and powerful in conducting hyperparameter optimization tasks.