View Project
Project Information
Project Title
Development of Subgradient-Based Optimization Algorithms for Interval-Valued Neural Networks
Activity Type
Research Assistant
Program Model
Plan A: GLIP
Eligibility: Exclusively for undergraduate or Master's degree students.
Duration: 1 semester or 3–4 months (depending on the international student's academic calendar)
University Financial Support:
- Tuition fees and other university fees
- On-campus accommodation fees (excluding electricity charges)
- Daily allowance for international students (240 Baht/day)
Eligibility: Exclusively for undergraduate or Master's degree students.
Duration: 1 semester or 3–4 months (depending on the international student's academic calendar)
University Financial Support:
- Tuition fees and other university fees
- On-campus accommodation fees (excluding electricity charges)
- Daily allowance for international students (240 Baht/day)
Duration
May 2027 - August 2027 (4 months)
Applicant Qualifications and Eligibility Requirements:
- Student Status: Applicants must maintain active student status at their home institution throughout the entire duration of the exchange program.
- Program Commitment & Return Requirements: Applicants must complete the full duration of the program and return to their home country on or before the official program end date, in strict compliance with immigration regulations.
- Scholarship Agreement: Applicants must acknowledge, accept, and comply with all rules, terms, and conditions of the scholarship.
- Overseas Health Insurance: Applicants must possess comprehensive overseas medical and health insurance coverage from their home country throughout their study, research, or training period in Thailand. The insurance benefits must explicitly cover sickness, injury, hospitalization expenses (both In-Patient and Out-Patient), medical and rescuer's expenses, personal liability, and death.
- Personal Financial Responsibility: Applicants must accept sole responsibility for all costs and expenses beyond the specified scholarship coverage.
- Schedule & Attendance for Course-Based Programs: For applicants enrolling in Course-Based Programs (Type: Registering in relevant courses), full-semester participation is required (November 16, 2026 – March 21, 2027). Participants must arrive at Naresuan University on November 12, 2026, and depart no later than March 22, 2027. Any changes to arrival or departure dates strictly require prior approval from the program coordinator and the university. (Note: For Internship and Research Assistantship programs, the duration and travel schedule will follow the specific timeline designated for each project).
Additional Project Qualifications:
Applicants who are interested in and have a strong background in optimizing machine learning algorithms and quantum machine learning will be preferred.
Project Status
Open
Activity Details
Overview
This research project aims to develop and analyze a subgradient-based optimization method for training neural networks with interval-valued inputs, utilizing the Hausdorff distance as the loss function. The study transitions from theoretical foundations to concrete algorithmic development by first formulating a convex model, such as interval-valued linear regression, and deriving explicit subgradient formulas. Following the establishment of rigorous theoretical convergence guarantees, the proposed algorithm will be computationally implemented and evaluated against both simulated and small-scale real-world datasets. The project will culminate in the potential extension to simple nonlinear neural networks and the preparation of a comprehensive academic manuscript summarizing the mathematical analyses and numerical findings.
The project timeline is structured into five phases:
Phase 1 (Weeks 1-3): Literature Review & Mathematical Formulation
Phase 2 (Weeks 4-7): Algorithm Derivation & Convergence Analysis
Phase 3 (Weeks 8-11): Computational Implementation & Simulation
Phase 4 (Weeks 12-14): Data Analysis & Optional Extension
Phase 5 (Weeks 15-16): Manuscript Preparation & Project Conclusion
The project timeline is structured into five phases:
Phase 1 (Weeks 1-3): Literature Review & Mathematical Formulation
Phase 2 (Weeks 4-7): Algorithm Derivation & Convergence Analysis
Phase 3 (Weeks 8-11): Computational Implementation & Simulation
Phase 4 (Weeks 12-14): Data Analysis & Optional Extension
Phase 5 (Weeks 15-16): Manuscript Preparation & Project Conclusion
Weekly Plan
| Week | Activity |
|---|---|
| Week 1 | Conduct a literature review on optimization techniques for interval-valued data and the properties of the Hausdorff loss function. |
| Week 2 | Formulate the basic mathematical model for interval-valued linear regression utilizing the Hausdorff distance. |
| Week 3 | Perform preliminary mathematical analysis and investigate the convexity and non-smooth characteristics of the objective function. |
| Week 4 | Derive explicit subgradient formulas for the formulated interval-valued convex model. |
| Week 5 | Verify the mathematical properties of the derived subgradients and establish the theoretical framework. |
| Week 6 | Conduct rigorous convergence analysis to establish theoretical guarantees for the proposed subgradient method. |
| Week 7 | Finalize mathematical proofs for the convergence theorems or establish empirical convergence bounds, and refine the optimization algorithm steps. |
| Week 8 | Set up the computational programming environment and implement the initial code for the subgradient algorithm. |
| Week 9 | Conduct numerical experiments using generated simulated interval-valued datasets to verify algorithmic correctness. |
| Week 10 | Debug, refine the codebase, and optimize computational performance based on initial simulation outcomes. |
| Week 11 | Evaluate the proposed algorithm's performance on selected small-scale real-world interval-valued datasets. |
| Week 12 | Analyze experimental results, compute performance metrics, and generate data visualizations for comparison. |
| Week 13 | Explore theoretical and practical extensions of the proposed method to a simple nonlinear neural network architecture. |
| Week 14 | Conduct preliminary computational experiments and performance evaluations on the extended nonlinear model. |
| Week 15 | Draft the research manuscript, integrating the mathematical formulations, convergence proofs, and numerical results. |
| Week 16 | Finalize the manuscript for potential academic submission, present project outcomes, and conclude the research program. |
Project Manager
Full Name (EN)
NARIN PETROT
Faculty / Institute
Faculty of Science
Email
narinp@nu.ac.th