Minjung Kang

Santiago Gonzalez, Alireza Amiri Bavandpour, Peter Ye et al.

As Large Language Models (LLMs) saturate elementary benchmarks, the research frontier has shifted from generation to the reliability of automated evaluation. We demonstrate that standard "LLM-as-a-Judge" protocols suffer from a systematic Alignment Gap when applied to upper-undergraduate to early graduate level mathematics. To quantify this, we introduce QEDBench, the first large-scale dual-rubric alignment benchmark to systematically measure alignment with human experts on university-level math proofs by contrasting course-specific rubrics against expert common knowledge criteria. By deploying a dual-evaluation matrix (7 judges x 5 solvers) against 1,000+ hours of human evaluation, we reveal that certain frontier evaluators like Claude Opus 4.5, DeepSeek-V3, Qwen 2.5 Max, and Llama 4 Maverick exhibit significant positive bias (up to +0.18, +0.20, +0.30, +0.36 mean score inflation, respectively). Furthermore, we uncover a critical reasoning gap in the discrete domain: while Gemini 3.0 Pro achieves state-of-the-art performance (0.91 average human evaluation score), other reasoning models like GPT-5 Pro and Claude Sonnet 4.5 see their performance significantly degrade in discrete domains. Specifically, their average human evaluation scores drop to 0.72 and 0.63 in Discrete Math, and to 0.74 and 0.50 in Graph Theory. In addition to these research results, we also release QEDBench as a public benchmark for evaluating and improving AI judges. Our benchmark is publicly published at https://github.com/qqliu/Yale-QEDBench.

Jaeyoung Cheong, Heejoon Lee, Minjung Kang

Recent researches on stock prediction using deep learning methods has been actively studied. This is the task to predict the movement of stock prices in the future based on historical trends. The approach to predicting the movement based solely on the pattern of the historical movement of it on charts, not on fundamental values, is called the Technical Analysis, which can be divided into univariate and multivariate methods in the regression task. According to the latter approach, it is important to select different factors well as inputs to enhance the performance of the model. Moreover, its performance can depend on which loss is used to train the model. However, most studies tend to focus on building the structures of models, not on how to select informative factors as inputs to train them. In this paper, we propose a method that can get better performance in terms of returns when selecting informative factors using the cointegration test and learning the model using quantile loss. We compare the two RNN variants with quantile loss with only five factors obtained through the cointegration test among the entire 15 stock index factors collected in the experiment. The Cumulative return and Sharpe ratio were used to evaluate the performance of trained models. Our experimental results show that our proposed method outperforms the other conventional approaches.