Multimodal semantic communication reduces uplink overhead by transmitting compact task-relevant representations. However, existing token compression methods either aggregate tokens using fixed structures or select tokens under fixed modality-wise budgets, without jointly adapting modality relevance and token utility to the task query. We propose hierarchical task-aware token compression (H-TATC), which allocates a shared non-text token budget across modalities and subsequently selects query-relevant tokens within each modality. On MUSIC-AVQA, H-TATC improves accuracy by 2.4 percentage points over TokCom-MLMs at B=64 and SNR =12 dB, and reaches 72.1% at B=128, only 0.7 points below the full-token reference, with negligible additional end-to-end latency.
This survey traces the evolution from LM-driven SemCom to TokenCom, and examines the token abstraction, the transmission techniques it requires, and two emerging paradigms, namely TokenCom for LM services and for embodied and agentic intelligence.
Omnimodal large language models (Omni-LLMs) encode audio and visual streams into temporally interleaved token sequences for multimodal reasoning. However, processing long audio-visual token sequences incurs substantial prefill costs. Existing compression methods have made progress, but often overlook temporal changes i...
Yu-Chen Deng, Zi-Dang Cai, Fei-Diao Yang et al.· 0 citations
Image-token compression must reduce computation without hurting the semantics. We introduce Varest, guided by semantic fungibility: tokens within a coherent semantic region can be summarized coarsely, while object boundaries and literal text require finer distinctions. Object and text localizers guide variable-resoluti...
Tian-Hao Fu, Austin S. Wang, Charles Chen et al.· 0 citations
Unified multimodal models combine understanding, generation, and editing within a single network, offering a promising foundation for versatile multimodal applications. However, growing multimodal contexts make KV cache storage and access increasingly costly. Existing KV cache compression methods are typically tailored...
Wan-Qi Yang, Yue-Xiao Ma, Mei Xie et al.· 1 citation
Video token compression must control both temporal coverage and task relevance, but sequential stages can conflict: a content-driven allocation stage may be overwritten by a global ranking that removes entire temporal positions. We identify this as an allocation conflict and propose TeQCo, a training-free framework tha...
Yu-Qi Li, Xiang-Long Wang, Neilier Rodrigues et al.· 0 citations
Efficient transmission of multimodal semantic features under strict physical-layer constraints is a key requirement for generative semantic communication. However, existing approaches often treat different semantic features uniformly despite their unequal impact on perceptual quality, which leads to inefficient resourc...
S. Kinoshita, Yue Yin, Mondher Bouazizi et al.· IEEE International Symposium...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
