Novel Transformer variants and non-Transformer architectures (e.g., Mamba, RWKV, State Space Models)
Unified representation, alignment, and fusion mechanisms for multimodal large models
Long-context modeling, extrapolation techniques, and efficient memory mechanisms
World models and the cognitive science foundations of language intelligence
Neuro-symbolic integration and structured reasoning
Emergent abilities and new discoveries in scaling laws of large models
Cross-lingual, low-resource language, and linguistic diversity modeling
◕ Track 2: Efficient Training, Inference, and Model Optimization
Efficient pre-training strategies, data curation, and curriculum learning
Model compression, quantization, pruning, and knowledge distillation
Inference acceleration, speculative decoding, and dynamic inference
Mixture-of-Experts (MoE), sparse activation, and conditional computation
Edge computing, on-device deployment, and mobile optimization of large models
Green AI, sustainable computing, and carbon footprint assessment
Continual learning and model editing