Lab Internship

for 2024 Winter

 LAB INTERNSHIP FOR THE 2024 WINTER
1 How to apply & Timeline

Submit your application form via Google Form https://forms.gle/f6eckwmiBUJvx7W18

Deadline of Submission: Nov. 13th (Wed), 2024
Selection: by Early-Dec, 2024

2 Internship Period

8 weeks during the 2024 Winter(2024.12.23(Mon) ~ 2025.02.21(Fri))

(May vary depending on the lab’s schedule)

3 Others
• This program is for undergraduate students and is open to students from other
departments and universities.
• Participating in the IE Frontier(Nov. 2025) is mandatory for KAIST ISE students
• If prior approval has been obtained from the professor and notified to the
department office, it is possible to register for IE495 Individual Study during the
internship period. 2
KAIST Department of Industrial & Systems Engineering 2
 LAB INTERNSHIP FOR THE 2024 WINTER
4 Credit recognition process for Graduation Research(IE490)
course
※ Only for the KAIST ISE Students
※ You can register for the Graduation Research(IE490) course only after
you have earned more than 94 credits by the 2024 Fall semester.

Timeline Contents
Evaluation for the internship result
After the Internship
(Pass / Fail) Supervisor
Notifying the Evaluation Result to the (Lab Internship)
After the Evaluation
department office
Course Registration &
Registering the IE490 course
Add / Drop Period Student
3 weeks before the end Submission of the Required Document -> Supervisor
of the semester - Bachelor Thesis(학사학위논문)
Submission of the Required Document
- Bachelor Thesis(학사학위논문) Supervisor
1 week before the end of
- Report on Evaluation of Bachelor -> Department
the semester
Dissertation(학사학위청구논문 심사결과 Office
보고서)

3
KAIST Department of Industrial & Systems Engineering 3
 ISE Labs and Research Topics(1/2)
No. of
Professor Lab Positions
Research Topic Details

Changhyun
COMET LAB 2 Large-scale Optimization Algorithms Link
Kwon
Woo Financial Artificial intelligence on financial engineering
4 Link
Chang Kim Engineering Lab. domain
Manufacturing &
Hyun-Jung
Service Systems 2 Scheduling of Manufacturing Systems with AI Link
Kim
Lab
Il-Chul
AAILab 4 Robot Vision-Language-Action LLM Link
Moon
Jinkyoo System Neural Combinatorial Optimization, Model-based
2 Link
Park Intelligence Lab Optimization
Data Science
Chanyoung Studying research papers and participating in the lab
and Artificial 4 Link
Park ’s research projects
Intelligence Lab

4
KAIST Department of Industrial & Systems Engineering 4
 ISE Labs and Research Topics(2/2)
No. of
Professor Lab Positions
Research Topic Details

Statistical
Hoseung Statistical change-point detection, AB testing,
Decision Making 2 Link
Song Association test, Clustering.
Lab.
Data Intelligence
Hwanjun
and System Lab 4 Trustworthy large language models Link
Song
(DISL)
Analytics/Decision making problems in healthcare
Taesik Lee CSD 2 Link
systems design
Young Jae Robot Orchestration in Manufacturing- Logistics
DAIM Lab 3 Link
Jang Automation and Digital Twin

5
KAIST Department of Industrial & Systems Engineering 5
 COMET LAB
• Topics in Computational Optimization Methods
• Decomposition-based large-scale optimization methods for
high-quality primal and dual solutions
• ML-based acceleration of large-scale optimization algorithms
• Applications in Urban Logistics and Urban Mobility
• Quantum Circuit Optimization Methods

• Course Requirements: OR1, Data Structure

• Experience with Julia, C/C++, or PyTorch is preferred

but optional.
• https://comet.kaist.ac.kr/join

6
KAIST Department of Industrial & Systems Engineering 6
 FINANCIAL ENGINEERING LAB.
• Topic: Artificial Intelligence for Financial
Engineering
• Research topics
• High Frequency Trading Strategy via Reinforcement Learning

• Asset-Liability Management via Reinforcement Learning

• Enterprise Risk Management via Multi-Agent Reinforcement Learning

• Life Time Financial Planning using Parametric Value Function Gradient

Approximation
• Market Analysis and Asset Management using Natural Language Processing

• Financial Instruments Recommender Systems based on Financial Big-Data

• Considerations
• Proficiency in coding and basic understanding on data analyzing is needed

• If you want to proceed with your own project, it will be organized as a team
and will be carried out under the guidance of assistants
7
KAIST Department of Industrial & Systems Engineering 7
 MANUFACTURING & SERVICE
SYSTEMS LAB
• Topic: Scheduling of Manufacturing Systems with AI
• Scheduling is the field of study concerned with the optimal
allocation or assignment of resources, over time, to a set of tasks
or activities.
• The types of resources and activities depend on the situations
• We will consider real-world scheduling problems for
manufacturing semiconductor, display, tire, home appliance, steel,
and so on.
• AI & RL can be used for scheduling.

8
KAIST Department of Industrial & Systems Engineering 8
 Compensation
AAILAB Monthly 600,000 won (2-Months)
Requiring 4 students

• Robot Multi-modal LLM

• Robot task planning has been approached by expanding LLM to multi-modal domain including
robot skillset actions.
• Its eventual goal is creating an embodied AI to manipulate a humanoid robot

Video Image Preferred

Text Tokens Learnable
Tokens Tokens Response
LLM
Linear Layer Prompt Dispreferred
Response
Video Q-Former
Relevant
Image Encoder Instruction Preferred
Action Seq. Reference
Response
LLM
Dispreferred
Query Action Response
Database

Video Frames Observation

Robot Observation
/ Human Instruction

Robot Task
Execution
9
KAIST Department of Industrial & Systems Engineering 9
 Compensation
AAILAB Monthly 600,000 won (2-Months)
Requiring 4 students

• Robot Multi-modal LLM

• Robot task planning has been approached by expanding LLM to multi-modal domain including
robot skillset actions.
• Its eventual goal is creating an embodied AI to manipulate a humanoid robot

• This research is a combination of Research Philosophy

• Deep generative model
Mathematically Principled
• Large Language model Approach
• (A mild) portion of robot HW
Theoretic
• This research provides some knowledge into Foundation
• Large-scale model finetuning

• Recent methods of generative modeling

Provable Learning and
Inference with Large-Scale
• Practical knowledge of coding in machine learning
Machine Learning Model
• Internship includes a lecture on the deep generative model

10
KAIST Department of Industrial & Systems Engineering 10
 SILAB (NCO)
• Topic: Neural Combinatorial Optimization (NCO)

• Internship Goals:
 Study recent NCO research of vehicle routing, graph covering, scheduling,
drug discovering in lens of reinforcement learning, generative modelling, and
discrete sampling.

 (Optional) Research for solving NCO problems using promising sampling

methods (e.g., GFlowNets, Diffusion Model, Langevin Monte Carlo) and
validate the idea on the demonstrated benchmark.

11
KAIST Department of Industrial & Systems Engineering 11
 SILAB (MBO)
• Topic: Model-Based Optimization (MBO)

• Internship Goals:
 Study recent MBO research on biological sequence design, neural
architecture design, robot morphology design in the lens of generative
modeling and active learning.

 (Optional) Research for achieving state-of-the-art performance on MBO

problems using promising methods (e.g., Diffusion Model, Bayesian
Optimization) and validate the idea on the demonstrated benchmark.

12
KAIST Department of Industrial & Systems Engineering 12
 SILAB (REQUIREMENTS)
• This requirement gives strong plus for the project
• Familiar with programming in Python.

• Knowledge for ML/DL/RL.

• (Optional) Knowledge for recent promising sampling-based generative

models (e.g., flow models, diffusion models).
• (Optional) Familiar with “PyTorch” or “JAX”, the autograd-based deep
learning framework.

• Interns will be team with multiple graduate students who are

working on similar topics.

13
KAIST Department of Industrial & Systems Engineering 13
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (1)

 Topic: Deep Learning / Large language model for Tabular Data

 Motivation
 Traditional ML models like Boosting methods generally perform better than deep learning on tabular data

 Challenge
 Heterogeneity / Sparsity / Dependency on pre-processing / Order invariant / Lack of prior knowledge

14
KAIST Department of Industrial & Systems Engineering 14
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (2)

 Topic: Solving science problems with AI (AI4Science)

 Bioinformatics: Spatial transcriptomics (Multi-omics)
 Chemistry: Analysis of IR Spectrum
 Materials Science: Material discovery, Material property prediction

Spatial
transcriptomics
(Bioinformatics)

Analysis of
IR Spectrum
(Chemistry)

Input: IR Spectrum (sequence data) Deep Learning Model Output: Molecule Structure (graph or string)

https://arxiv.org/pdf/2307.08423.pdf
15
KAIST Department of Industrial & Systems Engineering 15
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (3)

 Topic: Developing a model capable of reasoning based on mathematical-logical intelligence

(Task: Chart Reasoning - AI that solves problems by analyzing charts)
 Motivation:
 Even large language models like GPT exhibit very low performance in the area of chart reasoning.
 The goal is to develop a Vision-Language Model that can simultaneously understand both charts
(image) and questions (text).

Input (chart+question) Output (reasoning)

16
KAIST Department of Industrial & Systems Engineering 16
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (4)

 Topic: Generate a scene graph that stores memory about an image

 Given an Image $, generate a scene graph G !% &, ', (, ) *

V: Node O: Object class

maximize P G I ! P V I P E V, I P O, R V, E, I#
E: Edge R: Predicate class

Object Detection + Relatedness Prediction + Entity Classification + Predicate Classification

Image Captioning Image Generation

17
KAIST Department of Industrial & Systems Engineering 17
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (5)

 Topic: Constructing Dynamic Financial Knowledge Graph

 Motivation
 Stock market requires real-time analysis and decision-making in a rapidly changing data environment.
 Through real-time knowledge graphs, dynamic data can be effectively managed and utilized.
 Knowledge graphs can visualize the complex relationships in stock market data and update in real-time
to reflect the latest information.

18
KAIST Department of Industrial & Systems Engineering 18
 DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB (6)

 Topic: Explainable AI (Explainable Graph Neural Network)

 Task: What factors led the company to default? Analyzing Risk propagation.

 Motivation
 Studying explainable AI is essential to build trust and transparency in AI systems, allowing users to
understand and interpret the decisions made by complex models.

19
KAIST Department of Industrial & Systems Engineering 19
DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB
Tentative Schedule (Depends on progress)
 Study & implement research papers published by our group (Week 1~8)
 Paper seminar / week (2 papers per week)
 Checking progress and Q&A
 Implement & reproduce each paper
 1:1 meeting available upon request

 Participate in a project (Week 4~)

 Projects
 1. Deep Learning / Large language model for Tabular Data
 2. Solving science problems with AI (AI4Science)
 3. Developing a model capable of reasoning based on mathematical-logical intelligence
 4. Generate a scene graph that stores memory about an image
 5. Constructing Dynamic Financial Knowledge Graph
 6. Explainable AI (Explainable Graph Neural Network)
 … many others
 You can participate in the project you wish to participate (with graduate students)

20
KAIST Department of Industrial & Systems Engineering 20
DATA SCIENCE & ARTIFICIAL INTELLIGENCE LAB
• Requirement
• Familiar with programming in Python
• Knowledge in machine learning and deep learning
• Experience in deep learning libraries such as PyTorch, Tensorflow,
Keras is a strong plus
• Interns will team with graduate students
• If you have any topic you would like to work on, write about it in the
application

21
KAIST Department of Industrial & Systems Engineering 21
 STATISTICAL DECISION MAKING LAB.
• Topic: Statistical approaches to decision making
problems.

• Change-point detection, AB testing, Association

testing.

• Clustering, Community detection.

• Experience and study how statistics can be applied to

diverse decision making problems.

22
KAIST Department of Industrial & Systems Engineering 22
 DATA INTELLIGENCE AND SYSTEM LAB
• Topic: Trustworthy Large Language Models

• [1] Evaluation of large language models

• LLM benchmarking, automated evaluation

• [2] Reliable natural language inference with focus on AI Alignment

• Reliable LLM-based summarization & QA

• [3] Retrieval-augmented generation

• Document retrieval, query re-writing, evidence mapping

• [4] LLM-agent memory bank

• Management of LLM’s memory as an effective agent

• Course Requirements:
• Machine learning, Deep learning, Data science etc

23
KAIST Department of Industrial & Systems Engineering 23
 CSD (PROF. TAESIK LEE)
• Topic: Analytics/Decision making problems in healthcare
systems design

• Critical review of healthcare service resource capacity

planning studies in Korea

• Participation in lab projects in the area of healthcare

systems design and operation

• 과목요건: ISE 필수과목 중 6개 이상 이수

24
KAIST Department of Industrial & Systems Engineering 24
 DAIM LAB
• Topic: Robot Orchestration in Manufacturing-Logistics
Automation and Digital Twin

• Research 1: Large-scale robot-based automation

methods for manufacturing-logistics automation and
digital twin

• Research 2: Application of automation methodology

based on industry-academia projects

• Pre-requisite: IE251 Manufacturing Process Innovation

25
KAIST Department of Industrial & Systems Engineering 25