Chang-Hun Lee

Changhun Lee, Jungyu Jin, Taesu Kim et al.

Large language models (LLMs) with hundreds of billions of parameters require powerful server-grade GPUs for inference, limiting their practical deployment. To address this challenge, we introduce the outlier-aware weight quantization (OWQ) method, which aims to minimize LLM's footprint through low-precision representation. OWQ prioritizes a small subset of structured weights sensitive to quantization, storing them in high-precision, while applying highly tuned quantization to the remaining dense weights. This sensitivity-aware mixed-precision scheme reduces the quantization error notably, and extensive experiments demonstrate that 3.1-bit models using OWQ perform comparably to 4-bit models optimized by OPTQ. Furthermore, OWQ incorporates a parameter-efficient fine-tuning for task-specific adaptation, called weak column tuning (WCT), enabling accurate task-specific LLM adaptation with minimal memory overhead in the optimized format. OWQ represents a notable advancement in the flexibility, efficiency, and practicality of LLM optimization literature. The source code is available at https://github.com/xvyaward/owq

Chang-Hun Lee, Min-Guk Seo, Min-Jea Tahk et al.

A straight forward application of feedback linearization to the missile autopilot design for acceleration control may be limited due to the nonminimum characteristics and the model uncertainties. As a remedy, this paper presents a cascade structure of an acceleration controller based on approximate feedback linearization methodology with a time-delay adaptation scheme. The inner loop controller is constructed by applying feedback linearization to the approximate system which is a minimum phase system and provides the desired acceleration signal caused by the angle-of-attack. This controller is augmented by the time-delay adaptive law and the outer loop PI (proportional-integral) controller in order to adaptively compensate for feedback linearization error because of model uncertainty and in order to track the desired acceleration signal. The performance of the proposed method is examined through numerical simulations. Moreover, the proposed controller is tested by using an intercept scenario in 6DOF nonlinear simulations.

Chang-Hun Lee, Tae-Hun Kim, Min-Jea Tahk

This paper deals with a nonlinear adaptive autopilot design for agile missile systems. In advance of the autopilot design, an investigation of the agile turn maneuver, based on the trajectory optimization, is performed to determine state behaviors during the agile turn phase. This investigation shows that there exist highly nonlinear, rapidly changing dynamics and aerodynamic uncertainties. To handle of these difficulties, we propose a longitudinal autopilot for angle-of-attack tracking based on backstepping control methodology in conjunction with the time-delay adaptation scheme.

Changhun Lee, Chiehyeon Lim

We study the theoretical aspects of Reinforced Language Models (RLMs) from a bi-objective optimization perspective. Specifically, we consider the RLMs as a Pareto optimization problem that maximizes the two conflicting objectives, i.e., reward objective and likelihood objectives, simultaneously. Our main contribution consists of three parts. First, we establish the theoretical foundations of RLM as a Pareto optimization problem by presenting Reward Upper BOund (RUBO) and Pareto optimality. Our theoretical outcomes are supported by not only deductive proofs but also empirical results. Second, we propose Reward Dropout, a simple yet powerful method that guarantees to improve a bi-objective optimization of RLM. Lastly, we demonstrate that the Reward Dropout is consistently effective across five benchmark datasets and four benchmark LLMs, meaning that the Reward Dropout significantly improves the optimization performance of RLMs.