Siqi Fan, Xiusheng Huang, Yiqun Yao, Xuezhi Fang, Kang Liu, Peng Han, Shuo Shang, Aixin Sun, Yequan Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Accelerating large language model pre-training is a critical issue in present research. In this paper, we focus on speeding up pre-training by progressively growing from a small Transformer structure to a large one. There are two main research problems associated with progressive growth: determining the optimal growth schedule, and designing efficient growth operators. In terms of growth schedule, the impact of each single dimension on a schedule’s efficiency is underexplored by existing work. Regarding the growth operators, existing methods rely on the initialization of new weights to inherit knowledge, and achieve only non-strict function preservation, limiting further improvements on training dynamics. To address these issues, we propose Masked Structural Growth (MSG), including (i) growth schedules involving all possible dimensions and (ii) strictly function-preserving growth operators that is independent of the initialization of new weights. Experiments show that MSG is significantly faster than related work: we achieve up to 2.2x speedup in pre-training different types of language models while maintaining comparable or better downstream performances. Code is publicly available at https://github.com/cofe-ai/MSG.

Yiqun Yao, Michalis Papakostas, Mihai Burzo, Mohamed Abouelenien, Rada Mihalcea. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Multimodal sentiment analysis aims to detect and classify sentiment expressed in multimodal data. Research to date has focused on datasets with a large number of training samples, manual transcriptions, and nearly-balanced sentiment labels. However, data collection in real settings often leads to small datasets with noisy transcriptions and imbalanced label distributions, which are therefore significantly more challenging than in controlled settings. In this work, we introduce MORSE, a domain-specific dataset for MultimOdal sentiment analysis in Real-life SEttings. The dataset consists of 2,787 video clips extracted from 49 interviews with panelists in a product usage study, with each clip annotated for positive, negative, or neutral sentiment. The characteristics of MORSE include noisy transcriptions from raw videos, naturally imbalanced label distribution, and scarcity of minority labels. To address the challenging real-life settings in MORSE, we propose a novel two-step fine-tuning method for multimodal sentiment classification using transfer learning and the Transformer model architecture; our method starts with a pre-trained language model and one step of fine-tuning on the language modality, followed by the second step of joint fine-tuning that incorporates the visual and audio modalities. Experimental results show that while MORSE is challenging for various baseline models such as SVM and Transformer, our two-step fine-tuning method is able to capture the dataset characteristics and effectively address the challenges. Our method outperforms related work that uses both single and multiple modalities in the same transfer learning settings.

Visual reasoning is a special visual question answering problem that is multi-step and compositional by nature, and also requires intensive text-vision interactions.We propose CMM: Cascaded Mutual Modulation as a novel end-to-end visual reasoning model.CMM includes a multi-step comprehension process for both question and image.In each step, we use a Feature-wise Linear Modulation (FiLM) technique to enable textual/visual pipeline to mutually control each other.Experiments show that CMM significantly outperforms most related models, and reach stateof-the-arts on two visual reasoning benchmarks: CLEVR and NLVR, collected from both synthetic and natural languages.Ablation studies confirm that both our multistep framework and our visual-guided language modulation are critical to the task.Our code is available at https://github. com/FlamingHorizon/CMM-VR.

Recently, end-to-end memory networks have shown promising results on Question Answering task, which encode the past facts into an explicit memory and perform reasoning ability by making multiple computational steps on the memory. However, memory networks conduct the reasoning on sentence-level memory to output coarse semantic vectors and do not further take any attention mechanism to focus on words, which may lead to the model lose some detail information, especially when the answers are rare or unknown words. In this paper, we propose a novel Hierarchical Memory Networks, dubbed HMN. First, we encode the past facts into sentence-level memory and word-level memory respectively. Then, k-max pooling is exploited following reasoning module on the sentence-level memory to sample the k most relevant sentences to a question and feed these sentences into attention mechanism on the word-level memory to focus the words in the selected sentences. Finally, the prediction is jointly learned over the outputs of the sentence-level reasoning module and the word-level attention mechanism. The experimental results demonstrate that our approach successfully conducts answer selection on unknown words and achieves a better performance than memory networks.