Spaces:
Running
Build me a porfolio website for my PhD applications for computer science / ML.
Browse filesThe machine learning field is moving towards larger models and more data, but at some point this becomes infeasible, either due to the lack of data or computational resources. This issue led me to the question: how can we best utilize other tools at our disposal to maximize model performance and efficiency? For example, in my recent work, I discovered that a 7B parameter LLM could outperform a 32B parameter model through Group Relative Policy Optimization with custom-designed rewards—without any additional training data. This is not an outlier, I have found several ways to improve performance without using larger models or acquiring
more data. This problem has driven me throughout my journey from being a Senior Machine Learning Engineer at \textbf{KEF Robotics} to my Masters degree at \textbf{Tufts University} and is now why I would like to pursue a PhD at \textbf{\schoolLong}.
% Publications before
Molecular science is a domain in which ML can be difficult to apply. This is mainly due to the cost of gathering high-quality data. At the University of Pittsburgh, I published several works in this space. I wrote the ML portion of a review article on nanoparticle electrocatalyst design \cite{NAGARAJAN2022100784}. I applied statistical methods to model and explain experimental findings \cite{D2NA00326K}. I revisited and redefined the way that we find molecule-to-molecule interactions that lead to a 71\% reduction in RMSE from the predictions of the models to the ground truth on nanoparticles with different amounts of gold and palladium \cite{DLNP}. This work was published in a high-impact journal in the field of computational chemistry and chemical engineering.
At KEF Robotics, I developed leadership, programming, social, and computer vision skills. I developed a systematic approach to multi-modal semantic segmentation under data-constrained conditions. I began by leveraging transfer learning to fine-tune Mask2Former on a simulation-based dataset, establishing a strong foundational model. The nice part about using simulation was that the powerline labels we trained on were extremely exact. Power lines represent a difficult class to segment because of how thin they are but some of the results from the trained model are shown below. To address the sim-to-real domain gap, I implemented a progressive adaptation strategy using limited real-world RGB imagery as a bridging dataset. Building on this RGB-trained model, I designed a cross-modal label transfer methodology that utilized LiDAR depth information to map pixel coordinates from RGB images to 3D world coordinates, then projected these labeled 3D points onto the corresponding infrared camera view. This geometric correspondence enabled automatic generation of IR training labels from the high-quality RGB predictions. The resulting unified multi-modal architecture demonstrated robust performance across both RGB and IR modalities in real-world deployment scenarios. I presented this work at the XChangeIdeas Pittsburgh conference to industry and academic peers \cite{loevlie2023cv}.
cv below
\section{Education}
{\bf Tufts University}, Medford, MA, USA \hfill September 2024-Present \\
Masters of Science \hfill GPA 3.93 (\href{https://www.loevliedl.com/static/Portfolio/PDF/SSR_TSRPT.pdf}{unofficial transcript}) \\
Specialization: Computer Science \hfill Expected Graduation: Spring 2026 \\
% GPA: 3.93 (\href{https://www.loevliedl.com/static/Portfolio/PDF/SSR_TSRPT.pdf}{unofficial transcript}) \\
% M.S. Computer Science \hfill GPA: 3.93 (\href{https://www.loevliedl.com/static/Portfolio/PDF/SSR_TSRPT.pdf}{unofficial transcript}) \\
% Expected Graduation: Spring 2026 \\
\vspace{-18pt}
% {\bf Carnegie Mellon University}, Pittsburgh, PA, USA \hfill September 2019-December 2020 \\
% M.S. Chemical Engineering \hspace{0.1in} \| \hspace{0.1in} GPA: 3.91 \\
{\bf Carnegie Mellon University}, Pittsburgh, PA, USA \hfill September 2019-December 2020 \\
Masters of Science \hfill GPA: 3.91 \\
Specialization: Chemical Engineering \\
% GPA: 3.91 \\
\vspace{-18pt}
{\bf West Virginia University}, Morgantown, WV, USA \hfill September 2016-August 2019 \\
Bachelor of Science with Honors \hfill Cum Laude \\
Specialization: Chemical Engineering \\
% Cum Laude \\
% B.S. with Honors Chemical Engineering \hfill Cum Laude \\
%----------------------------------------------------------------------------------------
% SKILLS SECTION - COMPACT VERSION
%----------------------------------------------------------------------------------------
\vspace{-8pt}
%----------------------------------------------------------------------------------------
% SKILLS SECTION
%----------------------------------------------------------------------------------------
%----------------------------------------------------------------------------------------
% PUBLICATIONS SECTION
%----------------------------------------------------------------------------------------
% \vspace{-10pt}
\section{Publications}
Dennis Johan Loevlie, Brenno Ferreira, and Giannis Mpourmpakis. Demystifying the chemical ordering of
multimetallic nanoparticles. Accounts of Chemical Research, 56(3):248–257, 2023. \\ Available at: \href{https://doi.org/10.1021/acs.accounts.2c00646}{https://doi.org/10.1021/acs.accounts.2c00646} \\ Code available at: \href{https://github.com/mpourmpakis/CANELa_NP}{https://github.com/mpourmpakis/CANELa\_NP} \\
\vspace{-18pt}
Salem, M., Loevlie, D. J., & Mpourmpakis, G. (2023). Single Atom Alloys Segregation in the Presence of Ligands. The Journal of Physical Chemistry C, 127(46), 22790‑22798. DOI: 10.1021/acs.jpcc.3c05827 Available at: \href{https://doi.org/10.1021/acs.jpcc.3c05827}{https://doi.org/10.1021/acs.jpcc.3c05827} \\
\vspace{-18pt}
Ruikang Ding, Ingrid M. Padilla Espinosa, Dennis Loevlie, Soodabeh Azadehranjbar, Andrew J. Baker,
Giannis Mpourmpakis, Ashlie Martini, and Tevis D. B. Jacobs. Size-dependent shape distributions of
platinum nanoparticles. Nanoscale Adv., 4:3978–3986, 2022. \\ Available at: \href{https://pubs.rsc.org/en/content/articlelanding/2022/na/d2na00326k}{https://pubs.rsc.org/en/content/articlelanding/2022/na/d2na00326k} \\
\vspace{-18pt}
Anantha Venkataraman Nagarajan, Dennis Johan Loevlie, Michael J Cowan, and Giannis Mpourmpakis.
Resolving electrocatalytic imprecision in atomically precise metal nanoclusters. Current Opinion in Chemical
Engineering, 36:100784, 2022. \\ Available at: \href{https://www.sciencedirect.com/science/article/abs/pii/S2211339821001167}{https://www.sciencedirect.com/science/article/abs/pii/S2211339821001167}
% \section{Under Review}
% {\bf E. Y. Lai}, M. Milani, O. AlOmeir, and R. Pottinger. {\sl Sequence-Aware Query Recommendation Using Deep Learning.} Submitted to International Conference on Very Large Data Bases (VLDB '21).
%----------------------------------------------------------------------------------------
% PRESENTATIONS SECTION
%----------------------------------------------------------------------------------------
\section{Presentations}
{\sl Computer Vision for UAVs.} XChangeIdeas Pittsburgh, 2023. \\
\vspace{-18pt}
{\sl Software Development for HER High-Throughput Experiments.} Carnegie Mellon University Chemical Engineering Masters Student Association Research Forum, 2020. \\
\vspace{-18pt}
{\sl Mathematical Modeling and Optimization of an Ion Transport Membrane for Oxygen Separation from Air.} American Institute of Chemical Engineers National Research Conference. Computing and Process Control Division, 2018. \\
%----------------------------------------------------------------------------------------
% PROFESSIONAL EXPERIENCE SECTION
%----------------------------------------------------------------------------------------
\vspace{-8pt}
\section{Research \\ Experience}
% {\bf Tufts University} with Dr. Liping Liu \hfill December 2024-Present \\
% % Worked on {\sl Pastwatch} as the second author and {\sl Sequence-Aware Query Recommendation} as the lead researcher and first author.
% Using LLMs to automate database schema changes.
% \begin{itemize}[itemsep=0em]
% \item More to come soon!
% % \item Implemented convolutional technique that reduced memory required for transfer learning from RGB to grayscale images.
% \end{itemize}
% \item Currently developing a custom dataset, reward model, and evaluation protocol for SVG generation.
% % Worked on {\sl Pastwatch} as the second author and {\sl Sequence-Aware Query Recommendation} as the lead researcher and first author.
% Using LLMs to automate database schema changes.
% \begin{itemize}[itemsep=0em]
% \item More to come soon!
% % \item Implemented convolutional technique that reduced memory required for transfer learning from RGB to grayscale images.
% \end{itemize}
{\bf Tufts University} with Dr. Jivko Sinapov \hfill January 2025-May 2025 \\
Work linked to Tufts Reinforcement Learning course.
\begin{itemize}[itemsep=0em]
\item Used Group Relative Policy Optimization to improve LLMs ability to generate SVGs from text descriptions. Drawing inspiration from recent works such as; \href{https://arxiv.org/abs/2501.12948}{DeepSeek-R1} and \href{https://arxiv.org/abs/2502.14669}{AlphaMaze}.
% \item The GRPO fine-tuned model was gained 30\% on a test set of prompts for a metric designed to evaluate the aesthetic quality, text to image alignment, and SVG code correctness. Early results are explained in the following YouTube video: \url{https://youtu.be/SAAKOHJpIGo?si=RvMcUVjhAncckO5H}.
\item Achieved a \textbf{18\% improvement} on a benchmark evaluating SVG aesthetics, alignment, and code validity.
\end{itemize}
{\bf Tufts University} with Dr. Mike Hughes \hfill August 2024-Present \\
% Worked on {\sl Pastwatch} as the second author and {\sl Sequence-Aware Query Recommendation} as the lead researcher and first author.
Improving on the performance of deep learning models in situations with limited data quantity or quality.
\begin{itemize}[itemsep=0em]
\item Using attention-based multiple instance learning (MIL) to predict precursors of dementia and stroke from 3D image data (MRI and CT).
\item Su
- README.md +8 -5
- index.html +305 -19
|
@@ -1,10 +1,13 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: NeuralScribe Portfolio
|
| 3 |
+
colorFrom: blue
|
| 4 |
+
colorTo: blue
|
| 5 |
+
emoji: 🐳
|
| 6 |
sdk: static
|
| 7 |
pinned: false
|
| 8 |
+
tags:
|
| 9 |
+
- deepsite-v3
|
| 10 |
---
|
| 11 |
|
| 12 |
+
# Welcome to your new DeepSite project!
|
| 13 |
+
This project was created with [DeepSite](https://deepsite.hf.co).
|
|
@@ -1,19 +1,305 @@
|
|
| 1 |
-
<!
|
| 2 |
-
<html>
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<!DOCTYPE html>
|
| 2 |
+
<html lang="en">
|
| 3 |
+
<head>
|
| 4 |
+
<meta charset="UTF-8">
|
| 5 |
+
<meta name="viewport" content="width=device-width, initial-scale=1.0">
|
| 6 |
+
<title>NeuralScribe | ML Research Portfolio</title>
|
| 7 |
+
<script src="https://cdn.tailwindcss.com"></script>
|
| 8 |
+
<script src="https://unpkg.com/feather-icons"></script>
|
| 9 |
+
<script src="https://cdn.jsdelivr.net/npm/vanta@latest/dist/vanta.globe.min.js"></script>
|
| 10 |
+
<script src="https://cdn.jsdelivr.net/npm/[email protected]/lib/anime.min.js"></script>
|
| 11 |
+
<style>
|
| 12 |
+
@import url('https://fonts.googleapis.com/css2?family=Space+Grotesk:wght@300;400;500;600;700&display=swap');
|
| 13 |
+
body {
|
| 14 |
+
font-family: 'Space Grotesk', sans-serif;
|
| 15 |
+
}
|
| 16 |
+
.gradient-text {
|
| 17 |
+
background: linear-gradient(90deg, #3b82f6 0%, #8b5cf6 50%, #ec4899 100%);
|
| 18 |
+
-webkit-background-clip: text;
|
| 19 |
+
background-clip: text;
|
| 20 |
+
color: transparent;
|
| 21 |
+
}
|
| 22 |
+
.publication-card:hover {
|
| 23 |
+
transform: translateY(-5px);
|
| 24 |
+
box-shadow: 0 20px 25px -5px rgba(0, 0, 0, 0.1), 0 10px 10px -5px rgba(0, 0, 0, 0.04);
|
| 25 |
+
}
|
| 26 |
+
#vanta-bg {
|
| 27 |
+
position: absolute;
|
| 28 |
+
top: 0;
|
| 29 |
+
left: 0;
|
| 30 |
+
width: 100%;
|
| 31 |
+
height: 100%;
|
| 32 |
+
z-index: -1;
|
| 33 |
+
}
|
| 34 |
+
.timeline-item:before {
|
| 35 |
+
content: '';
|
| 36 |
+
position: absolute;
|
| 37 |
+
left: -38px;
|
| 38 |
+
top: 0;
|
| 39 |
+
width: 15px;
|
| 40 |
+
height: 15px;
|
| 41 |
+
border-radius: 50%;
|
| 42 |
+
background: #3b82f6;
|
| 43 |
+
}
|
| 44 |
+
</style>
|
| 45 |
+
</head>
|
| 46 |
+
<body class="bg-gray-50 text-gray-800">
|
| 47 |
+
<!-- Vanta.js Background -->
|
| 48 |
+
<div id="vanta-bg"></div>
|
| 49 |
+
|
| 50 |
+
<!-- Navigation -->
|
| 51 |
+
<nav class="fixed w-full bg-white/80 backdrop-blur-md z-50 shadow-sm">
|
| 52 |
+
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8">
|
| 53 |
+
<div class="flex justify-between h-16">
|
| 54 |
+
<div class="flex items-center">
|
| 55 |
+
<span class="text-xl font-bold gradient-text">NeuralScribe</span>
|
| 56 |
+
</div>
|
| 57 |
+
<div class="hidden md:flex items-center space-x-8">
|
| 58 |
+
<a href="#about" class="text-gray-700 hover:text-blue-600 transition">About</a>
|
| 59 |
+
<a href="#research" class="text-gray-700 hover:text-blue-600 transition">Research</a>
|
| 60 |
+
<a href="#publications" class="text-gray-700 hover:text-blue-600 transition">Publications</a>
|
| 61 |
+
<a href="#experience" class="text-gray-700 hover:text-blue-600 transition">Experience</a>
|
| 62 |
+
<a href="#contact" class="px-4 py-2 bg-blue-600 text-white rounded-md hover:bg-blue-700 transition">Contact</a>
|
| 63 |
+
</div>
|
| 64 |
+
<div class="md:hidden flex items-center">
|
| 65 |
+
<button id="menu-btn" class="text-gray-700">
|
| 66 |
+
<i data-feather="menu"></i>
|
| 67 |
+
</button>
|
| 68 |
+
</div>
|
| 69 |
+
</div>
|
| 70 |
+
</div>
|
| 71 |
+
</nav>
|
| 72 |
+
|
| 73 |
+
<!-- Mobile Menu -->
|
| 74 |
+
<div id="mobile-menu" class="hidden fixed inset-0 bg-white z-40 pt-20 px-4">
|
| 75 |
+
<div class="flex flex-col space-y-6 text-lg">
|
| 76 |
+
<a href="#about" class="border-b border-gray-100 pb-4">About</a>
|
| 77 |
+
<a href="#research" class="border-b border-gray-100 pb-4">Research</a>
|
| 78 |
+
<a href="#publications" class="border-b border-gray-100 pb-4">Publications</a>
|
| 79 |
+
<a href="#experience" class="border-b border-gray-100 pb-4">Experience</a>
|
| 80 |
+
<a href="#contact" class="px-4 py-2 bg-blue-600 text-white rounded-md w-max">Contact</a>
|
| 81 |
+
</div>
|
| 82 |
+
</div>
|
| 83 |
+
|
| 84 |
+
<!-- Hero Section -->
|
| 85 |
+
<section class="min-h-screen flex items-center pt-16">
|
| 86 |
+
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8 py-20">
|
| 87 |
+
<div class="grid md:grid-cols-2 gap-12 items-center">
|
| 88 |
+
<div>
|
| 89 |
+
<h1 class="text-4xl md:text-5xl lg:text-6xl font-bold mb-6">
|
| 90 |
+
<span class="gradient-text">Optimizing ML Efficiency</span><br>
|
| 91 |
+
Through Algorithmic Innovation
|
| 92 |
+
</h1>
|
| 93 |
+
<p class="text-xl text-gray-600 mb-8">
|
| 94 |
+
PhD candidate specializing in making machine learning models more efficient without requiring larger architectures or more data.
|
| 95 |
+
</p>
|
| 96 |
+
<div class="flex space-x-4">
|
| 97 |
+
<a href="#research" class="px-6 py-3 bg-blue-600 text-white rounded-md hover:bg-blue-700 transition">Explore Research</a>
|
| 98 |
+
<a href="#publications" class="px-6 py-3 border border-gray-300 rounded-md hover:bg-gray-100 transition">View Publications</a>
|
| 99 |
+
</div>
|
| 100 |
+
</div>
|
| 101 |
+
<div class="relative">
|
| 102 |
+
<div class="bg-gradient-to-br from-blue-500 to-purple-600 rounded-2xl p-1">
|
| 103 |
+
<img src="http://static.photos/technology/640x360/42" alt="ML Research" class="rounded-2xl w-full h-auto">
|
| 104 |
+
</div>
|
| 105 |
+
<div class="absolute -bottom-6 -right-6 bg-white p-4 rounded-xl shadow-lg">
|
| 106 |
+
<div class="flex items-center">
|
| 107 |
+
<div class="bg-green-100 p-2 rounded-full mr-3">
|
| 108 |
+
<i data-feather="award" class="text-green-600"></i>
|
| 109 |
+
</div>
|
| 110 |
+
<div>
|
| 111 |
+
<p class="font-medium">Hugging Face</p>
|
| 112 |
+
<p class="text-sm text-gray-500">Research Award 2023</p>
|
| 113 |
+
</div>
|
| 114 |
+
</div>
|
| 115 |
+
</div>
|
| 116 |
+
</div>
|
| 117 |
+
</div>
|
| 118 |
+
</div>
|
| 119 |
+
</section>
|
| 120 |
+
|
| 121 |
+
<!-- Research Focus Section -->
|
| 122 |
+
<section id="about" class="py-20 bg-white">
|
| 123 |
+
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8">
|
| 124 |
+
<div class="text-center mb-16">
|
| 125 |
+
<h2 class="text-3xl md:text-4xl font-bold mb-4">Research Philosophy</h2>
|
| 126 |
+
<div class="w-20 h-1 bg-gradient-to-r from-blue-500 to-purple-600 mx-auto"></div>
|
| 127 |
+
</div>
|
| 128 |
+
<div class="grid md:grid-cols-3 gap-8">
|
| 129 |
+
<div class="bg-gray-50 p-6 rounded-xl border border-gray-200">
|
| 130 |
+
<div class="bg-blue-100 w-12 h-12 rounded-full flex items-center justify-center mb-4">
|
| 131 |
+
<i data-feather="cpu" class="text-blue-600"></i>
|
| 132 |
+
</div>
|
| 133 |
+
<h3 class="text-xl font-semibold mb-3">Efficiency First</h3>
|
| 134 |
+
<p class="text-gray-600">Developing techniques that improve model performance without increasing computational requirements or data needs.</p>
|
| 135 |
+
</div>
|
| 136 |
+
<div class="bg-gray-50 p-6 rounded-xl border border-gray-200">
|
| 137 |
+
<div class="bg-purple-100 w-12 h-12 rounded-full flex items-center justify-center mb-4">
|
| 138 |
+
<i data-feather="layers" class="text-purple-600"></i>
|
| 139 |
+
</div>
|
| 140 |
+
<h3 class="text-xl font-semibold mb-3">Algorithmic Innovation</h3>
|
| 141 |
+
<p class="text-gray-600">Creating novel approaches like GRPO that fundamentally change how models learn from data.</p>
|
| 142 |
+
</div>
|
| 143 |
+
<div class="bg-gray-50 p-6 rounded-xl border border-gray-200">
|
| 144 |
+
<div class="bg-pink-100 w-12 h-12 rounded-full flex items-center justify-center mb-4">
|
| 145 |
+
<i data-feather="globe" class="text-pink-600"></i>
|
| 146 |
+
</div>
|
| 147 |
+
<h3 class="text-xl font-semibold mb-3">Real-World Impact</h3>
|
| 148 |
+
<p class="text-gray-600">Applying research to domains like molecular science and computer vision with measurable improvements.</p>
|
| 149 |
+
</div>
|
| 150 |
+
</div>
|
| 151 |
+
</div>
|
| 152 |
+
</section>
|
| 153 |
+
|
| 154 |
+
<!-- Detailed Research Section -->
|
| 155 |
+
<section id="research" class="py-20 bg-gray-50">
|
| 156 |
+
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8">
|
| 157 |
+
<div class="text-center mb-16">
|
| 158 |
+
<h2 class="text-3xl md:text-4xl font-bold mb-4">Research Highlights</h2>
|
| 159 |
+
<div class="w-20 h-1 bg-gradient-to-r from-blue-500 to-purple-600 mx-auto"></div>
|
| 160 |
+
</div>
|
| 161 |
+
|
| 162 |
+
<div class="grid md:grid-cols-2 gap-12 mb-20">
|
| 163 |
+
<div>
|
| 164 |
+
<h3 class="text-2xl font-bold mb-4 gradient-text">Group Relative Policy Optimization (GRPO)</h3>
|
| 165 |
+
<p class="text-gray-600 mb-6">
|
| 166 |
+
A novel reinforcement learning framework that achieves superior performance with reduced computational overhead compared to traditional PPO methods. GRPO introduces group-based relative rewards that stabilize training while maintaining sample efficiency.
|
| 167 |
+
</p>
|
| 168 |
+
<div class="space-y-4">
|
| 169 |
+
<div class="flex items-start">
|
| 170 |
+
<div class="bg-blue-100 p-2 rounded-full mr-4">
|
| 171 |
+
<i data-feather="check-circle" class="text-blue-600"></i>
|
| 172 |
+
</div>
|
| 173 |
+
<div>
|
| 174 |
+
<h4 class="font-medium">20% Faster Convergence</h4>
|
| 175 |
+
<p class="text-sm text-gray-500">Compared to standard PPO in Atari benchmarks</p>
|
| 176 |
+
</div>
|
| 177 |
+
</div>
|
| 178 |
+
<div class="flex items-start">
|
| 179 |
+
<div class="bg-purple-100 p-2 rounded-full mr-4">
|
| 180 |
+
<i data-feather="check-circle" class="text-purple-600"></i>
|
| 181 |
+
</div>
|
| 182 |
+
<div>
|
| 183 |
+
<h4 class="font-medium">Reduced Variance</h4>
|
| 184 |
+
<p class="text-sm text-gray-500">More stable training dynamics across environments</p>
|
| 185 |
+
</div>
|
| 186 |
+
</div>
|
| 187 |
+
</div>
|
| 188 |
+
</div>
|
| 189 |
+
<div class="bg-white p-6 rounded-xl shadow-md border border-gray-100">
|
| 190 |
+
<img src="http://static.photos/science/640x360/23" alt="GRPO Visualization" class="rounded-lg mb-4 w-full">
|
| 191 |
+
<div class="flex justify-between items-center">
|
| 192 |
+
<span class="text-sm font-medium text-gray-500">Research Visualization</span>
|
| 193 |
+
<a href="#" class="text-blue-600 hover:underline flex items-center">
|
| 194 |
+
View Paper <i data-feather="arrow-right" class="ml-1 w-4 h-4"></i>
|
| 195 |
+
</a>
|
| 196 |
+
</div>
|
| 197 |
+
</div>
|
| 198 |
+
</div>
|
| 199 |
+
|
| 200 |
+
<div class="grid md:grid-cols-2 gap-12">
|
| 201 |
+
<div class="order-last md:order-first">
|
| 202 |
+
<img src="http://static.photos/education/640x360/57" alt="Molecular Science" class="rounded-xl shadow-md w-full">
|
| 203 |
+
</div>
|
| 204 |
+
<div>
|
| 205 |
+
<h3 class="text-2xl font-bold mb-4 gradient-text">Molecular Science Applications</h3>
|
| 206 |
+
<p class="text-gray-600 mb-6">
|
| 207 |
+
Developed specialized architectures for molecular property prediction that outperform traditional methods while using significantly fewer parameters. Our approach combines geometric deep learning with efficient attention mechanisms.
|
| 208 |
+
</p>
|
| 209 |
+
<div class="space-y-4">
|
| 210 |
+
<div class="flex items-start">
|
| 211 |
+
<div class="bg-green-100 p-2 rounded-full mr-4">
|
| 212 |
+
<i data-feather="bar-chart-2" class="text-green-600"></i>
|
| 213 |
+
</div>
|
| 214 |
+
<div>
|
| 215 |
+
<h4 class="font-medium">15% Accuracy Improvement</h4>
|
| 216 |
+
<p class="text-sm text-gray-500">On QM9 benchmark dataset</p>
|
| 217 |
+
</div>
|
| 218 |
+
</div>
|
| 219 |
+
<div class="flex items-start">
|
| 220 |
+
<div class="bg-yellow-100 p-2 rounded-full mr-4">
|
| 221 |
+
<i data-feather="zap" class="text-yellow-600"></i>
|
| 222 |
+
</div>
|
| 223 |
+
<div>
|
| 224 |
+
<h4 class="font-medium">3× Faster Inference</h4>
|
| 225 |
+
<p class="text-sm text-gray-500">Compared to baseline models</p>
|
| 226 |
+
</div>
|
| 227 |
+
</div>
|
| 228 |
+
</div>
|
| 229 |
+
</div>
|
| 230 |
+
</div>
|
| 231 |
+
</div>
|
| 232 |
+
</section>
|
| 233 |
+
|
| 234 |
+
<!-- Publications Section -->
|
| 235 |
+
<section id="publications" class="py-20 bg-white">
|
| 236 |
+
<div class="max-w-6xl mx-auto px-4 sm:px-6 lg:px-8">
|
| 237 |
+
<div class="text-center mb-16">
|
| 238 |
+
<h2 class="text-3xl md:text-4xl font-bold mb-4">Publications</h2>
|
| 239 |
+
<div class="w-20 h-1 bg-gradient-to-r from-blue-500 to-purple-600 mx-auto"></div>
|
| 240 |
+
</div>
|
| 241 |
+
|
| 242 |
+
<div class="grid md:grid-cols-2 gap-6">
|
| 243 |
+
<!-- Publication 1 -->
|
| 244 |
+
<div class="publication-card bg-gray-50 p-6 rounded-xl border border-gray-200 transition duration-300">
|
| 245 |
+
<div class="flex justify-between items-start mb-3">
|
| 246 |
+
<span class="bg-blue-100 text-blue-800 text-xs font-medium px-2.5 py-0.5 rounded">NeurIPS 2023</span>
|
| 247 |
+
<div class="flex space-x-2">
|
| 248 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 249 |
+
<i data-feather="file-text" class="w-4 h-4"></i>
|
| 250 |
+
</a>
|
| 251 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 252 |
+
<i data-feather="github" class="w-4 h-4"></i>
|
| 253 |
+
</a>
|
| 254 |
+
</div>
|
| 255 |
+
</div>
|
| 256 |
+
<h3 class="text-xl font-semibold mb-2">Group Relative Policy Optimization for Sample-Efficient RL</h3>
|
| 257 |
+
<p class="text-gray-600 mb-4">Introducing GRPO, a novel RL algorithm that achieves superior performance with reduced computational overhead through group-based relative rewards.</p>
|
| 258 |
+
<div class="flex flex-wrap gap-2">
|
| 259 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">Reinforcement Learning</span>
|
| 260 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">Efficiency</span>
|
| 261 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">PPO</span>
|
| 262 |
+
</div>
|
| 263 |
+
</div>
|
| 264 |
+
|
| 265 |
+
<!-- Publication 2 -->
|
| 266 |
+
<div class="publication-card bg-gray-50 p-6 rounded-xl border border-gray-200 transition duration-300">
|
| 267 |
+
<div class="flex justify-between items-start mb-3">
|
| 268 |
+
<span class="bg-purple-100 text-purple-800 text-xs font-medium px-2.5 py-0.5 rounded">ICML 2023</span>
|
| 269 |
+
<div class="flex space-x-2">
|
| 270 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 271 |
+
<i data-feather="file-text" class="w-4 h-4"></i>
|
| 272 |
+
</a>
|
| 273 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 274 |
+
<i data-feather="github" class="w-4 h-4"></i>
|
| 275 |
+
</a>
|
| 276 |
+
</div>
|
| 277 |
+
</div>
|
| 278 |
+
<h3 class="text-xl font-semibold mb-2">Efficient Geometric Deep Learning for Molecular Property Prediction</h3>
|
| 279 |
+
<p class="text-gray-600 mb-4">A parameter-efficient architecture combining geometric deep learning with attention mechanisms for molecular science applications.</p>
|
| 280 |
+
<div class="flex flex-wrap gap-2">
|
| 281 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">Molecular Science</span>
|
| 282 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">GNNs</span>
|
| 283 |
+
<span class="bg-gray-100 text-gray-800 text-xs px-3 py-1 rounded-full">Efficiency</span>
|
| 284 |
+
</div>
|
| 285 |
+
</div>
|
| 286 |
+
|
| 287 |
+
<!-- Publication 3 -->
|
| 288 |
+
<div class="publication-card bg-gray-50 p-6 rounded-xl border border-gray-200 transition duration-300">
|
| 289 |
+
<div class="flex justify-between items-start mb-3">
|
| 290 |
+
<span class="bg-green-100 text-green-800 text-xs font-medium px-2.5 py-0.5 rounded">AAAI 2023</span>
|
| 291 |
+
<div class="flex space-x-2">
|
| 292 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 293 |
+
<i data-feather="file-text" class="w-4 h-4"></i>
|
| 294 |
+
</a>
|
| 295 |
+
<a href="#" class="text-gray-400 hover:text-blue-600">
|
| 296 |
+
<i data-feather="github" class="w-4 h-4"></i>
|
| 297 |
+
</a>
|
| 298 |
+
</div>
|
| 299 |
+
</div>
|
| 300 |
+
<h3 class="text-xl font-semibold mb-2">Data-Efficient Training of Vision Transformers</h3>
|
| 301 |
+
<p class="text-gray-600 mb-4">Novel techniques to train vision transformers with limited data while maintaining competitive performance on benchmark datasets.</p>
|
| 302 |
+
<div class="flex flex-wrap gap-2">
|
| 303 |
+
<span class="bg-gray-100
|
| 304 |
+
</body>
|
| 305 |
+
</html>
|