Vision-Language Models (VLMs) are powerful yet computationally intensive for widespread practical deployments. To address such challenge without costly re-training, post-training acceleration techniques like quantization and token reduction are extensively explored. However, current acceleration evaluations primarily target minimal overall performance degradation, overlooking a crucial question: does the accelerated model still give the same answers to the same questions as it did before acceleration? This is vital for stability-centered industrial applications where consistently correct answers for specific, known situations are paramount, such as in AI-based disease diagnosis. We systematically investigate this for accelerated VLMs, testing four leading models (LLaVA-1.5, LLaVA-Next, Qwen2-VL, Qwen2.5-VL) with eight acceleration methods on ten multi-modal benchmarks. Our findings are stark: despite minimal aggregate performance drops, accelerated models changed original answers up to 20% of the time. Critically, up to 6.5% of these changes converted correct answers to incorrect. Input perturbations magnified these inconsistencies, and the trend is confirmed by case studies with the medical VLM LLaVA-Med. This research reveals a significant oversight in VLM acceleration, stressing an urgent need for instance-level stability checks to ensure trustworthy real-world deployment.

Multi-modal Large Language Models (MLLMs) have achieved remarkable success by integrating visual and textual modalities. However, they incur significant computational overhead due to the large number of vision tokens processed, limiting their practicality in resource-constrained environments. We introduce Language-Guided Vision Token Pruning (LVPruning) for MLLMs, an effective yet simple method that significantly reduces the computational burden while preserving model performance. LVPruning employs cross-attention modules to compute the importance of vision tokens based on their interaction with language tokens, determining which to prune. Importantly, LVPruning can be integrated without modifying the original MLLM parameters, which makes LVPruning simple to apply or remove. Our experiments show that LVPruning can effectively reduce up to 90% of vision tokens by the middle layer of LLaVA-1.5, resulting in a 62.1% decrease in inference Tera Floating-Point Operations Per Second (TFLOPs), with an average performance loss of just 0.45% across nine multi-modal benchmarks.

Computational linguistics (CL) has witnessed tremendous advancements in recent years, with models such as large language models demonstrating exceptional performance in various natural language processing tasks. These advancements highlight their potential to help understand brain language processing, especially through the lens of brain encoding and decoding. Brain encoding involves the mapping of linguistic stimuli to brain activity, while brain decoding is the process of reconstructing linguistic stimuli from observed brain activities. CL models that excel at capturing and manipulating linguistic features are crucial for mapping linguistic stimuli to brain activities and vice versa. Brain encoding and decoding have vast applications, from enhancing human-computer interaction to developing assistive technologies for individuals with communication impairments. This tutorial will focus on elucidating how computational linguistics can facilitate brain encoding and decoding. We will delve into the principles and practices of using computational linguistics methods for brain encoding and decoding. We will also discuss the challenges and future directions of brain encoding and decoding. Through this tutorial, we aim to provide a comprehensive and informative overview of the intersection between computational linguistics and cognitive neuroscience, inspiring future research in this exciting and rapidly evolving field.

The CALCULUS project, drawing on human capabilities of imagination and commonsense for natural language understanding (NLU), aims to advance machine-based NLU by integrating traditional AI concepts with contemporary machine learning techniques. It focuses on developing anticipatory event representations from both textual and visual data, connecting language structure to visual spatial organization and incorporating broad knowledge domains. Through testing these models in NLU tasks and evaluating their ability to predict untrained spatial and temporal details using real-world metrics, CALCULUS employs machine learning methods, including Bayesian techniques and neural networks, especially in data-sparse scenarios. The project’s culmination involves creating demonstrators that transform written stories into dynamic videos, showcasing the interdisciplinary expertise of the project leader in natural language processing, language and visual data analysis, information retrieval, and machine learning, all vital for the project’s achievements. In the CALCULUS project, our exploration of machine translation extends beyond the conventional text-to-text framework. We are broadening the horizons of machine translation by delving into the essence of transforming the formats of data distribution while keeping the meaning. This innovative approach involves converting information from one modality into another, transcending traditional linguistic boundaries. Our project includes novel work on translating text into images and videos, brain signals into images and videos.

Argument structure learning (ASL) entails predicting relations between arguments. Because it can structure a document to facilitate its understanding, it has been widely applied in many fields (medical, commercial, and scientific domains). Despite its broad utilization, ASL remains a challenging task because it involves examining the complex relationships between the sentences in a potentially unstructured discourse. To resolve this problem, we have developed a simple yet effective approach called Dual-tower Multi-scale cOnvolution neural Network (DMON) for the ASL task. Specifically, we organize arguments into a relationship matrix that together with the argument embeddings forms a relationship tensor and design a mechanism to capture relations with contextual arguments. Experimental results on three different-domain argument mining datasets demonstrate that our framework outperforms state-of-the-art models. We will release the code after paper acceptance.

Decoding continuous language from brain activity is a formidable yet promising field of research. It is particularly significant for aiding people with speech disabilities to communicate through brain signals. This field addresses the complex task of mapping brain signals to text. The previous best attempt reverse-engineered this process in an indirect way: it began by learning to encode brain activity from text and then guided text generation by aligning with predicted brain responses. In contrast, we propose a simple yet effective method that guides text reconstruction by directly comparing them with the predicted text embeddings mapped from brain activities. Comprehensive experiments reveal that our method significantly outperforms the current state-of-the-art model, showing average improvements of 77% and 54% on BLEU and METEOR scores. We further validate the proposed modules through detailed ablation studies and case analyses and highlight a critical correlation: the more precisely we map brain activities to text embeddings, the better the text reconstruction results. Such insight can simplify the task of reconstructing language from brain activities for future work, emphasizing the importance of improving brain-to-text-embedding mapping techniques.

Accumulating knowledge to tackle new tasks without necessarily forgetting the old ones is a hallmark of human-like intelligence. But the current dominant paradigm of machine learning is still to train a model that works well on static datasets. When learning tasks in a stream where data distribution may fluctuate, fitting on new tasks often leads to forgetting on the previous ones. We propose a simple yet effective framework that continually learns natural language understanding tasks with one model. Our framework distills knowledge and replays experience from previous tasks when fitting on a new task, thus named DnR (distill and replay). The framework is based on language models and can be smoothly built with different language model architectures. Experimental results demonstrate that DnR outperfoms previous state-of-the-art models in continually learning tasks of the same type but from different domains, as well as tasks of different types. With the distillation method, we further show that it’s possible for DnR to incrementally compress the model size while still outperforming most of the baselines. We hope that DnR could promote the empirical application of continual language learning, and contribute to building human-level language intelligence minimally bothered by catastrophic forgetting.

Distributional semantic models (DSMs) generally require sufficient examples for a word to learn a high quality representation. This is in stark contrast with human who can guess the meaning of a word from one or a few referents only. In this paper, we propose Mem2Vec, a memory based embedding learning method capable of acquiring high quality word representations from fairly limited context. Our method directly adapts the representations produced by a DSM with a longterm memory to guide its guess of a novel word. Based on a pre-trained embedding space, the proposed method delivers impressive performance on two challenging few-shot word similarity tasks. Embeddings learned with our method also lead to considerable improvements over strong baselines on NER and sentiment classification.