Although deep learning has substantially advanced speech separation in recent years, most existing studies continue to prioritize separation quality while overlooking computational efficiency, an essential factor for low-latency speech processing in real-time applications. In this paper, we propose SepPrune, the first structured pruning framework specifically designed to compress deep speech separation models and reduce their computational cost. SepPrune begins by analyzing the computational structure of a given model to identify layers with the highest computational burden. It then introduces a differentiable masking strategy to enable gradient-driven channel selection. Based on the learned masks, SepPrune prunes redundant channels and fine-tunes the remaining parameters to recover performance. Extensive experiments demonstrate that this learnable pruning paradigm yields substantial advantages for channel pruning in speech separation models, outperforming existing methods. Notably, a model pruned with SepPrune can recover 85% of the performance of a pre-trained model (trained over hundreds of epochs) with only one epoch of fine-tuning, and achieves convergence 36x faster than training from scratch.

Modern large vision-language models (LVLMs) convert each input image into a large set of tokens that far outnumber the text tokens. Although this improves visual perception, it also introduces severe image token redundancy. Because image tokens contain sparse information, many contribute little to reasoning but greatly increase inference cost. Recent image token pruning methods address this issue by identifying important tokens and removing the rest. These methods improve efficiency with only small performance drops. However, most of them focus on single-image tasks and overlook multimodal in-context learning (ICL), where redundancy is higher and efficiency is more important. Redundant tokens weaken the advantage of multimodal ICL for rapid domain adaptation and lead to unstable performance. When existing pruning methods are applied in this setting, they cause large accuracy drops, which exposes a clear gap and the need for new approaches. To address this, we propose Contextually Adaptive Token Pruning (CATP), a training-free pruning method designed for multimodal ICL. CATP uses two stages of progressive pruning that fully reflect the complex cross-modal interactions in the input sequence. After removing 77.8% of the image tokens, CATP achieves an average performance gain of 0.6% over the vanilla model on four LVLMs and eight benchmarks, clearly outperforming all baselines. At the same time, it improves efficiency by reducing inference latency by an average of 10.78%. CATP strengthens the practical value of multimodal ICL and lays the foundation for future progress in interleaved image-text settings.

Electrocardiogram (ECG) is a common diagnostic indicator of heart disease in hospitals. Because of the low price and noninvasiveness of ECG diagnosis, it is widely used for prescreening and physical examination of heart diseases. In several studies on ECG analysis, only rough diagnoses are made to determine whether ECGs are abnormal or on a few kinds of ECG. In actual scenarios, doctors must analyze ECG samples in detail, which is a multilabel classification problem. Herein, we propose Hygeia, a multilabel deep learning-based ECG classification method that can analyze and classify 55 types of ECG. First, a guidance model is constructed to transform the multilabel classification problem into multiple interrelated two-classification models. This method ensures the good performance of each ECG analysis model, and the relationship between various types of ECG can be used in the analysis. The imbalance of samples in ECG datasets makes it difficult to analyze abnormal ECGs with high sensitivity and accuracy. We used data generation and mixed sampling methods for 11 ECG types with imbalanced problems to improve the average accuracy, sensitivity, F1 value, and accuracy from 87.74%, 43.11%, 0.3929, and 0.3929, to 92.68%, 96.92, 0.9287, and 99.47%, respectively. The average accuracy, sensitivity, F1 value, and accuracy of 44 of the 55 tags of the abnormal ECG analysis model are 99.69%, 95.81%, 0.9758, and 99.72%, respectively.

We aim at solving the problem of predicting people's ideology, or political tendency. We estimate it by using Twitter data, and formalize it as a classification problem. Ideology-detection has long been a challenging yet important problem. Certain groups, such as the policy makers, rely on it to make wise decisions. Back in the old days when labor-intensive survey-studies were needed to collect public opinions, analyzing ordinary citizens' political tendencies was uneasy. The rise of social medias, such as Twitter, has enabled us to gather ordinary citizen's data easily. However, the incompleteness of the labels and the features in social network datasets is tricky, not to mention the enormous data size and the heterogeneousity. The data differ dramatically from many commonly-used datasets, thus brings unique challenges. In our work, first we built our own datasets from Twitter. Next, we proposed TIMME, a multi-task multi-relational embedding model, that works efficiently on sparsely-labeled heterogeneous real-world dataset. It could also handle the incompleteness of the input features. Experimental results showed that TIMME is overall better than the state-of-the-art models for ideology detection on Twitter. Our findings include: links can lead to good classification outcomes without text; conservative voice is under-represented on Twitter; follow is the most important relation to predict ideology; retweet and mention enhance a higher chance of like, etc. Last but not least, TIMME could be extended to other datasets and tasks in theory.

The Massive Open Online Course (MOOC) is based on the Connectionism Theory and the Open Pedagogy of Network Learning, realizing the large-scale open sharing of high-quality education resources via the Internet. MOOC is conducive to the multi-dimension and personalized teaching. However, compared with the mode of traditional classroom teaching, there are also insurmountable drawbacks of MOOC to be addressed. Especially, it is difficult for educators to know the true states of students in real-time when they are learning online, and control the quality of teaching and learning. In this paper, we carried on the thorough research to the arousal degree of learner's brain, and proposed the MOOC-oriented EEG-based arousal of brain detection and adjustment scheme, and constructed the prototype system named Megrez. Megrez can analyze the brain arousal with the EEG data of learners wearing portable EEG detection devices. The energy value of β wave of EEG is selected as the main feature of brain arousal. After preprocessing the EEG data, the SVM classifier is trained, which classification performance is proved satisfactory. When a learner's brain is in a low arousal state, Megrez automatically plays the appropriate music to improve the arousal of brain. The experimental results show that Megrez can effectively detect the arousal of learner's brain, help learners avoid inefficient learning in the negative state, and can also adjust the state of the brain if necessary.

Semantic vectors, or language embeddings, are used in computational linguistics to represent language for a variety of machine related tasks including translation, speech to text, and natural language understanding. These semantic vectors have also been extensively studied in correlation with human brain data, showing evidence that the representation of language in the human brain can be modeled through these vectors with high correlation. Further, various attempts have been made to study how the human brain represents and understands music. For example, it has been shown that EEG data of subjects listening to music can be used for tempo detection and singer gender recognition. We propose studying the relationship between the EEG data of subjects listening to audio and the audio feature vectors modeled after the semantic vectors in computational linguistics. This could provide new insight into how the brain processes and understands music.

The brain is the locus of our language ability, and so brain images can be used to ground linguistic theories.Here we introduce Brain-Bench, a lightweight system for testing distributional models of word semantics.We compare the performance of several models, and show that the performance on brain-image tasks differs from the performance on behavioral tasks.We release our benchmark test as part of a web service.