model architectures
Model architectures refer to the specific design and structure of neural networks, including their layers, connectivity patterns, and operational mechanisms. Different architectures are suited for different types of tasks and data.
- Advancing Expert Specialization for Better MoE
- Advancing Expert Specialization for Better MoE
- Asymmetric Duos: Sidekicks Improve Uncertainty
- Benchmarking Spatiotemporal Reasoning in LLMs and Reasoning Models: Capabilities and Challenges
- Clean First, Align Later: Benchmarking Preference Data Cleaning for Reliable LLM Alignment
- CoUn: Empowering Machine Unlearning via Contrastive Learning
- Connecting Neural Models Latent Geometries with Relative Geodesic Representations
- FedWMSAM: Fast and Flat Federated Learning via Weighted Momentum and Sharpness-Aware Minimization
- Harnessing the Universal Geometry of Embeddings
- Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection
- Noise Injection Reveals Hidden Capabilities of Sandbagging Language Models
- RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video
- Scaling Law with Learning Rate Annealing
- TF-MAS: Training-free Mamba2 Architecture Search
- Taught Well Learned Ill: Towards Distillation-conditional Backdoor Attack
- The ML.ENERGY Benchmark: Toward Automated Inference Energy Measurement and Optimization
- Unleashing Foundation Vision Models: Adaptive Transfer for Diverse Data-Limited Scientific Domains