Update GSoC 2026 DeepLense project proposals

_gsocproposals/2026/proposal_DEEPLENSE1.md

@@ -1,53 +1,57 @@
 ---
-title: Foundation Model for Gravitational Lensing
+title: Agentic AI for Autonomous Gravitational Lensing Simulation Workflows
 layout: gsoc_proposal
 project: DEEPLENSE
 project size: 350hr
 year: 2026
 organization:
   - Alabama
   - MIT
-  - Florida
-  - PSL
+  - UAH
 ---
 
 ## Description
 
-Strong gravitational lensing is a powerful tool for studying dark matter and the large-scale structure of the universe. This project focuses on developing a vision foundation model specifically designed for lensing data, which can be fine-tuned for a variety of downstream tasks, including classification, super-resolution, regression, and lens finding. 
+Strong gravitational lensing is a powerful tool for studying dark matter and the large-scale structure of the universe. Current gravitational lensing simulation pipelines (e.g., [DeepLenseSim](https://github.com/mwt5345/DeepLenseSim) built on Lenstronomy) require substantial manual intervention: researchers must configure parameters, submit cluster jobs, validate outputs, retrain downstream ML models, and iterate on failures. This creates bottlenecks in large-scale dataset generation, limits exploration of parameter space, and consumes researcher time on engineering tasks rather than scientific analysis.
+This project proposes the development of an Agentic AI framework to autonomously orchestrate gravitational lensing simulation workflows. Unlike traditional automation scripts that follow rigid rules, agentic systems employ LLM-powered agents that can reason about scientific objectives, adapt to failures, and coordinate complex multi-step workflows with minimal human supervision.
 
-This project will explore different training strategies such as self-supervised learning, contrastive learning, or transformer-based models to learn meaningful representations of lensing images. By leveraging diverse datasets and training methodologies, the model will serve as a general-purpose backbone that can adapt to different astrophysical tasks while improving generalization across various observational conditions.
 
 ## Duration
 
-Total project length: 350 hours.
+Total project length: 175/350 hours.
 
 ## Difficulty level
 
 Advanced
 
 ## Task ideas
- * Develop a pre-training strategy for learning robust representations of gravitational lensing data.
- * Fine-tune the foundation model for multiple tasks such as classification, super-resolution, and regression.
- * Evaluate the model's performance on different astrophysical datasets and benchmark against traditional methods.
+ * Survey agent frameworks and orchestration patterns
+ * Design multi-agent system with specialized agents
+ * Integrate with DeepLenseSim and SLURM
+ * Evaluate automation gains and dataset quality
+
 
 ## Expected results
- *  A vision foundation model for gravitational lensing capable of being fine-tuned for various astrophysical tasks.
- *  Improved generalization and adaptability across different lensing datasets and observational setups.
+ * Working multi-agent system for autonomous workflow orchestration
+ * Significant reduction in manual intervention with adaptive behavior
+ * Benchmark study comparing against traditional pipelines
+
 
 ## Requirements
- * Python, PyTorch, experience with machine learning, and familiarity with computer vision techniques.
- * Understanding of self-supervised learning, representation learning, and deep learning architectures.
+ * Prior experience with LLMs and local inference tools such as Llama.cpp or Ollama
+ * Prior experience with agentic AI frameworks such as Orchestral AI or Pydantic AI 
+ * Prior experience with Reinforcement Learning (RL) and its application to AI agents, planning, or autonomous workflow optimization is highly preferred.
+ * Familiarity with gravitational lensing is preferred but not mandatory.
+
 
-<!---
 ## Test
-Please use this [link](https://docs.google.com/document/d/1a-5JiHph3K59gV3-kEZWzKYTFMvDeYiJvoE0U2I4x0w/edit?usp=sharing) to access the test for this project.
---->
+Please use this [link](https://docs.google.com/document/d/10APh49fvayGoSftzO4fGXs2HP3uvYSzG-fSrq4xHL1w/edit?usp=sharing) to access the test for this project.
 
 ## Mentors
   * [Michael Toomey](mailto:ml4-sci@cern.ch) (Massachusetts Institute of Technology)
   * [Sergei Gleyzer](mailto:ml4-sci@cern.ch) (University of Alabama)
-  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (University of Florida)
-  * [Anna Parul](mailto:ml4-sci@cern.ch) (Observatoire de Paris)
+  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (Independent Researcher)
+  * [Rajat Shinde](mailto:ml4-sci@cern.ch) (University of Alabama in Huntsville)
 
 
 Please **DO NOT** contact mentors directly by email. General questions can be directed to [ml4-sci@cern.ch](mailto:ml4-sci@cern.ch). To submit your proposal, CV, and test task solutions, please use [this Google form](https://forms.gle/SPXo8kSwHHptcBmk9).
@@ -56,5 +60,5 @@ Please **DO NOT** contact mentors directly by email. General questions can be di
 ## Links
   * [Paper 1](https://arxiv.org/abs/2008.12731)
   * [Paper 2](https://arxiv.org/abs/1909.07346)
-  * [Paper 3](https://arxiv.org/abs/2112.12121)
-
+  * [Paper 3](https://arxiv.org/pdf/2512.15867)
+

_gsocproposals/2026/proposal_DEEPLENSE2.md

@@ -7,8 +7,8 @@ year: 2026
 organization:
   - Alabama
   - MIT
-  - Florida
   - PSL
+  - UAH
 ---
 
 ## Description
@@ -25,20 +25,26 @@ Total project length: 175/350 hours.
 Advanced
 
 ## Task ideas
+ * Implement hybrid quantum-classical architectures (variational quantum circuits, quantum kernels) for dark matter substructure classification from strong lensing images.
+ * Benchmark quantum models against classical baselines and evaluate noise robustness under realistic NISQ conditions.
 
 ## Expected results
+ * Trained hybrid quantum-classical model for multi-class lensing classification.
+ * Benchmark of quantum vs. classical performance with analysis of near-term hardware feasibility.
 
 ## Requirements
  * Strong background in Machine Learning & Deep Learning.
  * Knowledge of Quantum Computing (VQAs, Quantum Optimization).
  * Proficiency in Python & Pennylane or Qiskit.
  * Ability to work independently on research projects.
 
+## Test
+Please use this [link](https://docs.google.com/document/d/10APh49fvayGoSftzO4fGXs2HP3uvYSzG-fSrq4xHL1w/edit?usp=sharing) to access the test for this project.
 
 ## Mentors
   * [Michael Toomey](mailto:ml4-sci@cern.ch) (Massachusetts Institute of Technology)
   * [Sergei Gleyzer](mailto:ml4-sci@cern.ch) (University of Alabama)
-  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (University of Florida)
+  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (Independent Researcher)
   * [Rajat Shinde](mailto:ml4-sci@cern.ch) (University of Alabama in Huntsville)
 
 

_gsocproposals/2026/proposal_DEEPLENSE3.md

@@ -1,53 +1,96 @@
 ---
-title: Physics Guided Machine Learning on Real Lensing Images 
+title: Neural Operators for Fast Simulation of Strong Gravitational Lensing
 layout: gsoc_proposal
 project: DEEPLENSE
-project size: 175hr/350hr
+project size: 350hr
 year: 2026
 organization:
   - Alabama
   - MIT
+  - IIT(ISM) Dhanbad
 ---
 
 ## Description
 
-This project focuses on developing a Physics-Informed Neural Network (PINN) framework for analyzing real strong gravitational lensing datasets to study dark matter distribution. Strong gravitational lensing, a key prediction of general relativity, occurs when a massive galaxy or cluster bends light from a background source, creating arcs or Einstein rings. Traditional algorithms struggle or fail entirely when applied to real lensing datasets due to observational complexities and noise. By leveraging PINNs, the project will integrate physical laws directly into the learning process, enhancing the accuracy and interpretability of dark matter inferences. The model will be trained on real lensing images, incorporating observational constraints to refine mass distribution estimates and improve the efficiency of dark matter studies.
+Strong gravitational lensing is a powerful probe of dark matter substructure and cosmological structure formation. Current simulation pipelines (e.g., Lenstronomy-based ray tracing) require solving the lens equation repeatedly for different mass distributions and source configurations. While accurate, these simulations are computationally expensive and limit large-scale parameter sweeps, uncertainty quantification, and real-time inference.
+This project proposes the development of a Neural Operator framework to learn the functional mapping between mass distributions and lensed images. Unlike traditional neural networks that operate on finite-dimensional vectors, neural operators learn mappings between infinite-dimensional function spaces. This makes them particularly well-suited for approximating solutions of physical systems governed by partial differential equations.
+The goal is to train a neural operator that directly maps:
+  * Lens mass distribution → Lensed image
+  or
+  * (Mass distribution, source light profile) → Observed lensed image
+
+
+Such a model would serve as a fast surrogate simulator capable of producing high-fidelity lensing outputs at a fraction of the computational cost of traditional ray-tracing solvers.
+This would represent the first exploration of neural operators within the ML4SCI DeepLense ecosystem.
+
 
 ## Duration
 
 Total project length: 175/350 hours.
 
 ## Difficulty level
 
-Intermediate/Advanced
+Advanced
 
 ## Task ideas
- * Build various physics-informed neural network architectures that are endowed with known physics for real lensing datasets. 
- * Apply these models to study dark matter in strong lensing images in various contexts: classification, regression, anomaly detection, and more.
-
+1. Literature Study
+    1. Study Fourier Neural Operators (FNO)
+    2. Study DeepONets
+    3. Review operator learning in physical systems
+2. Neural Operator Implementation
+    1. Implement Fourier Neural Operator for 2D fields
+    2. Compare with DeepONet-style architectures
+    3. Explore spectral vs spatial operator parameterizations
+3. Evaluation
+    1. Compare speed vs traditional solver
+    2. Measure pixel-wise reconstruction error
+    3. Evaluate preservation of physical invariants
+    4. Test generalization across:
+        1. Different mass profiles
+        2. Different redshifts
+        3. Different source morphologies
+4. Extensions (if time permits)
+    1. Conditional neural operators
+    2. Uncertainty-aware operator learning
+    3. Physics-informed operator constraints
+    4. Hybrid operator + diffusion refinement
+
+
+
+
 ## Expected results
- *  A more capable architecture that can operate on a wider variety of lensing images, including lensing images created with real galaxy datasets.
- *  Insight into the lensing systems, and their sub-structures.
+ * A neural operator capable of approximating the lensing simulation operator.
+ * Significant speedup compared to ray-tracing pipelines.
+ * Demonstration of generalization across varying astrophysical conditions.
+ * A benchmark study comparing neural operators with CNN-based surrogate models.
+
+
 
 ## Requirements
-Python, PyTorch, experience with machine learning, knowledge of computer vision techniques, familiarity with autoencoders.
+Python, PyTorch, experience with machine learning and deep learning.
+Partial understanding of:
+  * Representation learning
+  * Spectral methods
+  * Partial differential equations (basic familiarity)
+  * Scientific machine learning concepts
+Familiarity with gravitational lensing is preferred but not mandatory.
+
 
-<!---
 ## Test
-Please use this [link](https://docs.google.com/document/d/1a-5JiHph3K59gV3-kEZWzKYTFMvDeYiJvoE0U2I4x0w/edit?usp=sharing) to access the test for this project.
---->
+Please use this [link](https://docs.google.com/document/d/10APh49fvayGoSftzO4fGXs2HP3uvYSzG-fSrq4xHL1w/edit?usp=sharing) to access the test for this project.
 
 ## Mentors
   * [Michael Toomey](mailto:ml4-sci@cern.ch) (Massachusetts Institute of Technology)
   * [Sergei Gleyzer](mailto:ml4-sci@cern.ch) (University of Alabama)
+  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (Independent Researcher)
   * [Ashutosh Ojha](mailto:ml4-sci@cern.ch) (Indian Institute of Technology (Indian School of Mines), Dhanbad)
 
 
 Please **DO NOT** contact mentors directly by email. General questions can be directed to [ml4-sci@cern.ch](mailto:ml4-sci@cern.ch). To submit your proposal, CV, and test task solutions, please use [this Google form](https://forms.gle/SPXo8kSwHHptcBmk9).
 
 
 ## Links
-  * [Paper 1](https://arxiv.org/abs/2008.12731)
-  * [Paper 2](https://arxiv.org/abs/1909.07346)
-  * [Paper 3](https://ml4physicalsciences.github.io/2024/files/NeurIPS_ML4PS_2024_78.pdf)
-  * [Paper 4](https://ml4physicalsciences.github.io/2025/files/NeurIPS_ML4PS_2025_252.pdf)
+  * [Paper 1](https://arxiv.org/abs/2010.08895)
+  * [Paper 2](https://arxiv.org/abs/2108.08481)
+  * [Paper 3](https://arxiv.org/abs/1910.03193)
+

_gsocproposals/2026/proposal_DEEPLENSE4.md

@@ -1,8 +1,8 @@
 ---
-title: Gravitational Lens Finding 
+title: Foundation Model for Gravitational Lensing
 layout: gsoc_proposal
 project: DEEPLENSE
-project size: 175hr/350hr
+project size: 350hr
 year: 2026
 organization:
   - Alabama
@@ -12,43 +12,46 @@ organization:
 
 ## Description
 
-This project focuses on the task of lens finding in the currently available wide-field surveys (e.g., HSC-SSP). The expected number of strong lenses in the large surveys is significantly overpowered by the number of non-lensed objects, which leads to the high number of false positives in typical lens searches. 
+Strong gravitational lensing is a powerful tool for studying dark matter and the large-scale structure of the universe. This project focuses on developing a vision foundation model specifically designed for lensing data, which can be fine-tuned for a variety of downstream tasks, including classification, super-resolution, regression, and lens finding. 
 
-The goal of the project is to develop lens finding algorithms, apply them to the observational data, and  assess the limitations of the algorithms (for example, analyse the properties of the identified lens population and examine the typical contaminants).
+This project will explore different training strategies such as self-supervised learning, contrastive learning, or transformer-based models to learn meaningful representations of lensing images. By leveraging diverse datasets and training methodologies, the model will serve as a general-purpose backbone that can adapt to different astrophysical tasks while improving generalization across various observational conditions.
 
 ## Duration
 
-Total project length: 175/350 hours.
+Total project length: 350 hours.
 
 ## Difficulty level
-Intermediate/Advanced
+
+Advanced
 
 ## Task ideas
- * Start with architectures previously explored within the DeepLense project and optimise them for the lens finding task
- * Perform the lens search in the real observational data and analyse the properties of the detected lens candidates
- * Evaluate model performance on different surveys
+ * Develop a pre-training strategy for learning robust representations of gravitational lensing data.
+ * Fine-tune the foundation model for multiple tasks such as classification, super-resolution, and regression.
+ * Evaluate the model's performance on different astrophysical datasets and benchmark against traditional methods.
 
 ## Expected results
- *  Increase the number of known strong lenses.
- *  Insight into properties of the identified lens candidates.
-   
+ *  A vision foundation model for gravitational lensing capable of being fine-tuned for various astrophysical tasks.
+ *  Improved generalization and adaptability across different lensing datasets and observational setups.
+
 ## Requirements
-Python, PyTorch, experience with machine learning, familiarity with astrophysics datasets.
+ * Python, PyTorch, experience with machine learning, and familiarity with computer vision techniques.
+ * Understanding of self-supervised learning, representation learning, and deep learning architectures.
 
-<!---
 ## Test
-Please use this [link](https://docs.google.com/document/d/1a-5JiHph3K59gV3-kEZWzKYTFMvDeYiJvoE0U2I4x0w/edit?usp=sharing) to access the test for this project.
---->
+Please use this [link](https://docs.google.com/document/d/10APh49fvayGoSftzO4fGXs2HP3uvYSzG-fSrq4xHL1w/edit?usp=sharing) to access the test for this project.
 
 ## Mentors
-  * [Sergei Gleyzer](mailto:ml4-sci@cern.ch) (University of Alabama)
   * [Michael Toomey](mailto:ml4-sci@cern.ch) (Massachusetts Institute of Technology)
+  * [Sergei Gleyzer](mailto:ml4-sci@cern.ch) (University of Alabama)
+  * [Pranath Reddy](mailto:ml4-sci@cern.ch) (Independent Researcher)
   * [Anna Parul](mailto:ml4-sci@cern.ch) (Observatoire de Paris)
 
 
 Please **DO NOT** contact mentors directly by email. General questions can be directed to [ml4-sci@cern.ch](mailto:ml4-sci@cern.ch). To submit your proposal, CV, and test task solutions, please use [this Google form](https://forms.gle/SPXo8kSwHHptcBmk9).
 
 
 ## Links
-  * [Paper 1](https://arxiv.org/abs/1909.07346)
-  * [Paper 2](https://ml4physicalsciences.github.io/2024/files/NeurIPS_ML4PS_2024_107.pdf)
+  * [Paper 1](https://arxiv.org/abs/2008.12731)
+  * [Paper 2](https://arxiv.org/abs/1909.07346)
+  * [Paper 3](https://arxiv.org/abs/2112.12121)
+