Research

Paper reading, experiment tracking, and lab project records.

A conservative research workspace for interests, reading lists, logs, experiments, and future results.

Interests

Research interests to develop over time.

The goal is to connect reading, implementation, and experiments rather than listing directions without context.

Artificial IntelligenceComputer VisionScientific image parsingDeep LearningTransformer and CNN systemsOCC and 3D VisionGaussian SplattingResearch EngineeringScientific Computing

Reading List

Paper reading fields are ready for MDX migration later.

TitleVenueTopicStatusNotes

Attention Is All You Need

Vaswani et al. · 2017

NeurIPSTransformerReadingOpen

An Image is Worth 16x16 Words

Dosovitskiy et al. · 2021

ICLRVision TransformerTo ReadOpen

You Only Look Once: Unified Real-Time Object Detection

Redmon et al. · 2016

CVPRObject DetectionSummarizedOpen

3D Gaussian Splatting for Real-Time Radiance Field Rendering

Kerbl et al. · 2023

SIGGRAPH3D VisionTo ReadOpen

Research Logs

Mass spectra prototype retrospective

Mass Spectra Image Recognition

2026-06-04

Organized the project as a two-week prototype: rule-based peak extraction, browser-side demo, and CNN / Transformer upgrade path.

Blockers: Current public demo is still a lightweight rule-based system. Reliable chemical identification needs labeled spectra, trained weights, and spectral library validation.

Next: Prepare annotated examples, define peak-level labels, and compare rule outputs with future CNN / Transformer modules.

Set up research logging structure

Boxer Project

2026-05-30

Defined fields for weekly progress, blockers, experiment settings, and next actions.

Blockers: Need to connect logs with real datasets and lab discussion outcomes.

Next: Create the first real project note after the next meeting or experiment run.

CNN / Transformer reading pass

Vision Learning

2026-05-23

Collected key vocabulary around convolution, attention, inductive bias, and model scaling.

Blockers: Need controlled experiments instead of only reading notes.

Next: Run a small classification demo and compare training curves.

Object detection vocabulary map

Computer Vision Notes

2026-05-16

Organized IoU, AP, NMS, anchor-based and anchor-free detection concepts.

Blockers: Metrics need hands-on implementation to become reliable.

Next: Add a small metric calculation notebook.

Experiments

Experiment tracking for reproducible research practice.

Each row is designed to become a real experiment entry with config, dataset, result, and conclusion.

Mass spectra peak extraction baseline

dataset
Clean mass spectra screenshots and demo examples
model
Rule-based peak detector with future CNN / Transformer interfaces
config
Image preprocessing, plotting-area estimation, vertical peak detection, pixel-to-m/z conversion
result
Public browser demo can display peak candidates and rule-based fragment explanations
conclusion
A useful interpretable baseline for presentation; future reliability depends on annotated data and trained model weights.

CNN baseline sanity check

dataset
Small image classification dataset
model
Compact CNN
config
3 conv blocks, Adam optimizer, fixed seed, basic augmentation
result
Placeholder for first reproducible baseline
conclusion
Use this slot to compare future Transformer or detection experiments.

Detection metric notebook

dataset
Toy bounding-box examples
model
Metric-only implementation
config
IoU and AP calculation from simple annotations
result
Pending implementation
conclusion
Designed to make evaluation vocabulary concrete before running larger models.

Boxer Project prototype log

dataset
Lab dataset placeholder
model
Project-specific model placeholder
config
To be filled after project setup is finalized
result
Reserved for demo screenshots, tables, or qualitative results
conclusion
Keep results conservative and reproducible.

Lab Project

Boxer Project

In Progress

Background

A reserved lab project area for organizing real background, datasets, assumptions, and research questions as the work becomes clear.

Goal

Build a structured record of project goals, reproducible experiments, and technical decisions without inventing unsupported claims.

Current Progress

Initial documentation structure is prepared. Real milestones can be added after lab meetings, code runs, or experiment reports.

Technical Stack

Python, PyTorch, experiment tracking notes, Git, and future visualization or demo components.

Demo / Results

Reserved for screenshots, tables, qualitative results, and links to reproducible demos.

Future Work

Connect project logs to datasets, baseline experiments, ablation notes, and a readable final project page.