Recent & Upcoming Talks

In this session, I’m excited to take you on a deep dive into the cutting-edge world of artificial intelligence and computer vision, as we explore the process of training an AI model to tackle the classic challenge of “Where’s Waldo.” This isn’t just about spotting a red-and-white striped figure—it’s about unlocking the power of AI to recognize and understand intricate patterns in images. As we break down the complexities of image recognition, we’ll harness the latest techniques in deep learning, including the revolutionary Vision Transformers, to train a model that can pick out Waldo from a sea of distractions. This session will equip you with hands-on experience in AI model development, giving you the tools and knowledge to apply to a wide range of real-world problems in visual recognition.

In this quick pitch, I’m thrilled to introduce MeViT—a medium-resolution Vision Transformer developed for high-precision semantic segmentation of Landsat satellite imagery.

I recently had the opportunity to speak with high school students about ‘Career Paths for AI Research Scientists.’ During the talk, I shared my experiences as a postdoctoral researcher in AI, diving into the exciting world of artificial intelligence. I discussed the various career opportunities in the field, from academic research to industry roles, and highlighted how my own journey has shaped my understanding of Generative AI.

I had the opportunity to give a final orientation speech to the undergraduate students of the Department of Electrical and Computer Engineering at KMUTNB. The focus of my speech was on the transformative impact of AI, particularly highlighting the advancements in Large Language Models (LLMs) like ChatGPT. I discussed how these models have revolutionized natural language processing, enabling sophisticated interactions and problem-solving capabilities.

Geospatial Data Analytics involves analyzing spatial and geographical data to gain insights and make informed decisions. Using PySpark, this process is accelerated through distributed computing, enabling the handling of large datasets efficiently. Distributed Machine Learning models further enhance the analysis by providing scalable and robust predictions. Visualization tools like Looker Studio present the analyzed data in an interactive and comprehensible format, facilitating better decision-making and strategic planning. This combination of technologies allows for comprehensive geospatial data analysis, uncovering patterns and trends that drive actionable insights.

I was invited to Italy to present my research, “MARS, Mask Attention Refinement with Sequential Quadtree Nodes,” at the ICIAP 2023 Workshop. This prestigious conference, organized biennially by CVPL under the International Association for Pattern Recognition (IAPR), will unite experts to discuss advancements in car insurance and computer vision. My research addresses the challenge of accurately evaluating car damages using MARS, which enhances instance segmentation through self-attention mechanisms and quadtree transformers.

I was invited to teach a course on distributed machine learning to the Geo-Informatics and Space Technology Development Agency (GISTDA). The curriculum covered fundamental concepts of PySpark, basic deep learning techniques, and practical applications of distributed training using TensorFlow. I also emphasized methods for leveraging Multi-GPU setups and implementing distributed training strategies, particularly in the context of geospatial data analytics, equipping participants with the skills needed to handle large-scale machine learning tasks efficiently.

I was invited to speak at the “Achieve Data Science First Meet” for a MOOC student project event, where I highlighted the growing recognition of data science, AI, and machine learning’s importance across various industries. I advised that organizations, regardless of their size or sector, must effectively develop and implement data science capabilities to stay competitive in the era of big data, or risk falling behind.

This dissertation introduces a new architecture for remote sensing, featuring Global Convolutional Network (GCN), channel attention, domain-specific transfer learning, Feature Fusion (FF), and Depthwise Atrous Convolution (DA). Tests on Landsat-8 and ISPRS Vaihingen datasets show that this model significantly outperforms the baseline.

I was honored to be invited as the Head TA for the Data Science Pathway course organized by CHULA MOOC Achieve. In this role, I guided students through the practical applications of data science using RapidMiner and machine learning, along with Python programming. The curriculum encompassed both supervised and unsupervised learning labs, featuring tasks such as Customer Retention Prediction and Boston Housing Price Prediction, among others.

Hey everyone! I’m Teerapong Panboonyuen, but you can call me Kao. I’m passionate about creating impactful AI. While working on my Ph.D. in Computer Engineering at Chula, I’ve taught, tackled exciting data science projects, and assisted in machine learning classes. My goal? To elevate AI and leverage it for a better world. My journey has been a thrilling exploration of advanced technologies, from diffusion models and generative adversarial networks (GANs) to self-supervised learning and probabilistic modeling.