Skip to content
View psychofict's full-sized avatar

Block or report psychofict

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
psychofict/README.md

Ebenezer Tarubinga

Portfolio  ·  Google Scholar  ·  LinkedIn  ·  Website  ·  Email


I build AI systems that go from research to production. Currently an AI Research Engineer at Gractor, designing and deploying computer vision pipelines for smart city infrastructure. Previously published 2 first-author papers on semi-supervised semantic segmentation at Korea University's Pattern Recognition & ML Lab (MSc in AI, advised by Prof. Seong-Whan Lee), and before that shipped 10+ full-stack products as a software engineer.

I care about closing the gap between what works in a paper and what works at scale.

Publications

Paper Venue Links
CW-BASS: Confidence-Weighted Boundary Aware Learning for Semi-Supervised Semantic Segmentation IJCNN 2025 Paper · Code · Project Page
FARCLUSS: Fuzzy Adaptive Rebalancing and Contrastive Uncertainty Learning for Semi-Supervised Semantic Segmentation Neural Networks 2026 Paper · Code · Project Page

Both papers address the core challenge of learning with limited labels — reducing pseudo-label noise, handling class imbalance, and sharpening boundary predictions. CW-BASS achieves 65.9% mIoU on Cityscapes with only 100 labeled images. FARCLUSS introduces fuzzy pseudo-labeling and prototype-based contrastive regularization, setting new baselines on Pascal VOC and Cityscapes.

Selected Projects

Research

Project Area Links
Backdoor Attacks on CLIP (BadCLIP) Adversarial ML / Multimodal Code · Slides
Semantic-Aware Multi-Label Adversarial Attacks Adversarial ML Code · Slides
Semi-Supervised Segmentation Baselines (UniMatch + ST++) Semi-Supervised Learning Code · Slides
Monocular Depth Estimation (Depth Anything V2) 3D Vision / Autonomous Driving Code
Effective Context Length of LLMs (STRING) NLP / Transformers Code · Slides

Software

Project What it does Stack
AgriLet AI crop disease detection + agricultural education platform React, TypeScript, Express, Prisma
eLearn E-learning platform with 20+ subjects, analytics, progress tracking React, TypeScript, Express, Prisma
3D Digital Twin Real-time traffic monitoring with 3D visualization and edge AI React, Three.js, FastAPI

Skills

ML/CV: PyTorch · TensorFlow · OpenCV · CUDA · ONNX · MLflow · NVIDIA Jetson · Docker

Software: React · TypeScript · Node.js · Express · FastAPI · Three.js · Prisma · PostgreSQL

Languages: Python · C++ · C# · Java · Bash

Experience

AI Research Engineer · Gractor · Sept 2025 – present
Production CV systems for smart city AIoT; real-time inference on edge devices.

AI Research Engineer · Korea University · Aug 2023 – Aug 2025
2 first-author publications; segmentation models outperforming baselines by up to 25% mIoU; autonomous driving patent filed.

AI Software Engineer · GliTech · Jan 2019 – Jan 2021
10+ shipped fullstack products; led engineering for AI-powered applications serving hundreds of users.

Education

MSc in Artificial Intelligence · Korea University · 2023–2025
Global Korea Scholarship · Advisor: Prof. Seong-Whan Lee

Pinned Loading

  1. FARCLUSS FARCLUSS Public

    [Neural Networks 2026] Fuzzy Adaptive Rebalancing and Contrastive Uncertainty Learning for Semi-Supervised Semantic Segmentation

    Python 2

  2. CW-BASS CW-BASS Public

    [IJCNN 2025] 'CW-BASS: Confidence-Weighted Boundary Aware Learning for Semi-Supervised Semantic Segmentation'

    Python 5