Language networks are crucial in artificial intelligence, with the novel Mamba architecture significantly reducing computations and consumption compared to the traditional transformer network. However, a full-circuit implementation of the Mamba network has not been proposed due to the complexity of computations and data storage. Additionally, optimized hardware-aware parallel algorithms for Mamba inference in circuits remain undeveloped. This work addresses these challenges by presenting a memristor-based full-circuit implementation of the Mamba network and introducing a computing-in-memory parallel-aware algorithm tailored for circuit-level inference. The implementation includes: 1) Standard 1T1M memristor crossbar and depthwise separable convolution memristor crossbar for different convolutions. 2) Computing-in-memory implicit latent state circuits for the computation and transition of latent states. 3) Functional circuits for SiLU activation, RMS normalization, and multi-layer multiply-accumulate operations. 4) Optimized algorithm and circuit implementation for hardware-aware inference, achieving parallel scanning and hardware awareness in circuits. The proposed circuit enables analog signal computations and eliminates redundant analog-to-digital conversions and intermediate storage. A basic single-sentence generation task was simulated in PSPICE, validating the circuit’s correctness. Analyses of analog computation accuracy, circuit stability, and power consumption demonstrate the proposed circuit’s advantages, highlighting its potential as a fundamental module for large-scale circuit integration and complex text generation tasks.

Complementary semantic information from the text modality, obtained through runtime transcription, plays a crucial role in Speech Emotion Recognition (SER). However, it introduces additional computational overhead and potential errors. To address these issues, we propose the SDMMKD framework, which directly leverages multimodal knowledge without runtime transcription. Specifically, SDMMKD distills emotion knowledge at both the feature and logit levels from a pre-trained multimodal teacher during training. During inference, SDMMKD relies solely on speech signals to perform unimodal SER. Additionally, we utilize a Mamba block to enhance dynamic temporal features. Experimental results on the widely used IEMOCAP dataset demonstrate that our proposed SDMMKD framework outperforms state-of-the-art methods, achieving a WAR of 75.89% and a UAR of 77.37%.

In real-world physiological and psychological scenarios, there often exists a robust complementary correlation between audio and visual signals. Audio-Visual Event Localization (AVEL) aims to identify segments with Audio-Visual Events (AVEs) that contain both audio and visual tracks in unconstrained videos. Prior studies have predominantly focused on audio-visual cross-modal fusion methods, overlooking the fine-grained exploration of the cross-modal information fusion mechanism. Moreover, due to the inherent heterogeneity of multi-modal data, inevitable new noise is introduced during the audio-visual fusion process. To address these challenges, we propose a novel Cross-modal Contrastive Learning Network (CCLN) for AVEL, comprising a backbone network and a branch network. In the backbone network, drawing inspiration from physiological theories of sensory integration, we elucidate the process of audio-visual information fusion, interaction, and integration from an information-flow perspective. Notably, the Self-constrained Bi-modal Interaction (SBI) module is a bi-modal attention structure integrated with audio-visual fusion information, and through gated processing of the audio-visual correlation matrix, it effectively captures inter-modal correlation. The Foreground Event Enhancement (FEE) module emphasizes the significance of event-level boundaries by elongating the distance between scene events during training through adaptive weights. Furthermore, we introduce weak video-level labels to constrain the cross-modal semantic alignment of audio-visual events and design a weakly supervised cross-modal contrastive learning loss (WCCL Loss) function, which enhances the quality of fusion representation in the dual-branch contrastive learning framework. Extensive experiments conducted on the AVE dataset for both fully supervised and weakly supervised event localization, as well as Cross-Modal Localization (CML) tasks, demonstrate the superior performance of our model compared to state-of-the-art approaches.

Audio and vision are important senses for high-level cognition, and their special strong correlation makes audio-visual coding a crucial factor in many multimodal tasks. However, there are two challenges in audio-visual coding. First, the heterogeneity of multimodal data often leads to misalignment of cross-modal features under the same sample, which reduces their representation quality. Second, most self-supervised learning frameworks are constructed based on instance semantics, and the generated pseudo labels introduce additional classification noise. To address these challenges, we propose a Supervised Cross-modal Contrastive Learning Framework for Audio-Visual Coding (SCLAV). Our framework includes an audio-visual coding network composed of an inter-modal attention interaction module and an intra-modal self-integration module, which leverage multimodal complementary and hidden information for better representation. Additionally, we introduce a supervised cross-modal contrastive loss to minimize the distance between audio and vision features of the same instance, and use weak labels of multimodal data to eliminate the feature-oriented classification noise. Extensive experiments on the AVE and XD-Violence datasets demonstrate that SCLAV outperforms the state-of-the-art results, even with limited computational resources.