Limin Wang

Limin Wang, Masatoshi Hanai, Toyotaro Suzumura et al.

Revealing and analyzing the various properties of materials is an essential and critical issue in the development of materials, including batteries, semiconductors, catalysts, and pharmaceuticals. Traditionally, these properties have been determined through theoretical calculations and simulations. However, it is not practical to perform such calculations on every single candidate material. Recently, a combination method of the theoretical calculation and machine learning has emerged, that involves training machine learning models on a subset of theoretical calculation results to construct a surrogate model that can be applied to the remaining materials. On the other hand, a technique called pre-training is used to improve the accuracy of machine learning models. Pre-training involves training the model on pretext task, which is different from the target task, before training the model on the target task. This process aims to extract the input data features, stabilizing the learning process and improving its accuracy. However, in the case of molecular property prediction, there is a strong imbalance in the distribution of input data and features, which may lead to biased learning towards frequently occurring data during pre-training. In this study, we propose an effective pre-training method that addresses the imbalance in input data. We aim to improve the final accuracy by modifying the loss function of the existing representative pre-training method, node masking, to compensate the imbalance. We have investigated and assessed the impact of our proposed imbalance compensation on pre-training and the final prediction accuracy through experiments and evaluations using benchmark of molecular property prediction models.

Guo Chen, Yicheng Liu, Yifei Huang et al.

Most existing video understanding benchmarks for multimodal large language models (MLLMs) focus only on short videos. The limited number of benchmarks for long video understanding often rely solely on multiple-choice questions (MCQs). However, because of the inherent limitation of MCQ-based evaluation and the increasing reasoning ability of MLLMs, models can give the current answer purely by combining short video understanding with elimination, without genuinely understanding the video content. To address this gap, we introduce CG-Bench, a novel benchmark designed for clue-grounded question answering in long videos. CG-Bench emphasizes the model's ability to retrieve relevant clues for questions, enhancing evaluation credibility. It features 1,219 manually curated videos categorized by a granular system with 14 primary categories, 171 secondary categories, and 638 tertiary categories, making it the largest benchmark for long video analysis. The benchmark includes 12,129 QA pairs in three major question types: perception, reasoning, and hallucination. Compensating the drawbacks of pure MCQ-based evaluation, we design two novel clue-based evaluation methods: clue-grounded white box and black box evaluations, to assess whether the model generates answers based on the correct understanding of the video. We evaluate multiple closed-source and open-source MLLMs on CG-Bench. Results indicate that current models significantly underperform in understanding long videos compared to short ones, and a significant gap exists between open-source and commercial models. We hope CG-Bench can advance the development of more trustworthy and capable MLLMs for long video understanding. All annotations and video data are released at https://cg-bench.github.io/leaderboard/.

Jun Zhang, Desen Meng, Zhengming Zhang et al.

Despite the remarkable performance of multimodal large language models (MLLMs) across diverse tasks, the substantial training and inference costs impede their advancement. In this paper, we propose p-MoD, an efficient MLLM architecture that significantly reduces training and inference costs while maintaining model performance. The majority of computation in MLLMs stems from the overwhelming volume of vision tokens processed by the transformer-based LLM. Accordingly, we leverage the Mixture-of-Depths (MoD) mechanism, where each LLM layer selects essential vision tokens to process while skipping redundant ones. However, integrating MoD into MLLMs is non-trivial. To address the challenges of training and inference stability as well as limited training data, we adapt the MoD module with two novel designs: tanh-gated weight normalization (TanhNorm) and symmetric token reweighting (STRing). Moreover, we observe that vision tokens exhibit higher redundancy in deeper layers and thus design a progressive ratio decay (PRD) strategy, which gradually reduces the token retention ratio layer by layer, employing a shifted cosine schedule. This crucial design fully unleashes the potential of MoD, significantly boosting the efficiency and performance of our models. Extensive experiments on two baseline models across 15 benchmarks show that our model matches or even surpasses the performance of corresponding baselines, while requiring only 55.6% TFLOPs and 53.7% KV cache storage during inference, and 77.7% GPU hours during training.