AAAI 2026 accepted papers on Architectures and efficiency
197 of the 4,417 papers accepted at AAAI 2026 (4.5%) are on Architectures and efficiency, compared with 4.6% at AAAI 2025. They fall into 8 topics.
See AAAI 2026 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Attention and transformers · 55 papers
- Vision Transformers Are Circulant Attention Learners (5 citations)
- Parameterized Abstract Interpretation for Transformer Verification (2 citations)
- Transolver Is a Linear Transformer: Revisiting Physics-Attention Through the Lens of Linear Attention (2 citations)
- Aligning Attention with Human Rationales for Self-Explaining Hate Speech Detection (1 citation)
- Collaborative Transformers with Multi-Level Forensic Attention for Image Manipulation Localization (1 citation)
- Head-Aware KV Cache Compression for Efficient Visual Autoregressive Modeling (1 citation)
- Learning Spatial Decay for Vision Transformers (1 citation)
- Share Your Attention: Transformer Weight Sharing via Matrix-based Dictionary Learning (1 citation)
- SlimInfer: Accelerating Long-Context LLM Inference via Dynamic Token Pruning (1 citation)
- V-Pruner: A Fast and Globally-informed Token Pruning Framework for Vision Transformer (1 citation)
- Vision-MoR: Scaling Vision Transformer via Patch-Level Mixture-of-Recursions (1 citation)
- AMS-KV: Adaptive KV Caching in Multi-Scale Visual Autoregressive Transformers
and 43 more in the interactive map.
Pruning and efficient tuning · 46 papers
- SepPrune: Structured Pruning for Efficient Deep Speech Separation (11 citations)
- Small but Mighty: Dynamic Wavelet Expert-Guided Fine-Tuning of Large-Scale Models for Optical Remote Sensing Object Segmentation (3 citations)
- UNSEEN: Enhancing Dataset Pruning from a Generalization Perspective (2 citations)
- D2 Prune: Sparsifying Large Language Models via Dual Taylor Expansion and Attention Distribution Awareness (1 citation)
- DIET: Machine Unlearning on a Data-Diet (1 citation)
- Efficiently Seeking Flat Minima for Better Generalization in Fine-Tuning Large Language Models and Beyond (1 citation)
- Escaping Optimization Stagnation: Taking Steps Beyond Task Arithmetic via Difference Vectors (1 citation)
- Forgetting by Pruning: Data Deletion in Join Cardinality Estimation (1 citation)
- GateRA: Token-aware Modulation for Parameter-Efficient Fine-tuning (1 citation)
- Group Orthogonal Low-Rank Adaptation for RGB-T Tracking (1 citation)
- RoSA: Enhancing Parameter-Efficient Fine-Tuning via RoPE-aware Selective Adaptation in Large Language Models (1 citation)
- SALR: Sparsity-Aware Low-Rank Representation for Efficient Fine-Tuning of Large Language Models (1 citation)
and 34 more in the interactive map.
Quantization · 26 papers
- FlashSVD: Memory-Efficient Inference with Streaming for Low-Rank Models (6 citations)
- LampQ: Towards Accurate Layer-wise Mixed Precision Quantization for Vision Transformers (3 citations)
- DynaQuant: Dynamic Mixed-Precision Quantization for Learned Image Compression (1 citation)
- First-Order Error Matters: Accurate Compensation for Quantized Large Language Models (1 citation)
- HALO: Hardware-Aware Quantization with Low Critical-Path-Delay Weights for LLM Acceleration (1 citation)
- InfoQ: Mixed-Precision Quantization via Global Information Flow (1 citation)
- KVmix: Gradient-Based Layer Importance-Aware Mixed-Precision Quantization for KV Cache (1 citation)
- Mix-QSAM2: Mixed-Precision Quantization for High Fidelity Segmentation in Resource Constrained Scenarios (1 citation)
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks?
- BitDP: Ultra-low-bit Communication for Data Parallelism in LLM Training
- Double Rounding: Nearly Lossless Adaptive Bit Switching for QAT
- DQT: Dynamic Quantization Training via Dequantization-Free Nested Integer Arithmetic
and 14 more in the interactive map.
Spiking networks · 23 papers
- Spikingformer: A Key Foundation Model for Spiking Neural Networks (7 citations)
- Exploring the Potentials of Spiking Neural Networks for Image Deraining (2 citations)
- SpikingIR: A Novel Converted Spiking Neural Network for Efficient Image Restoration (2 citations)
- Temporal Dynamics Enhancer for Directly Trained Spiking Object Detectors (2 citations)
- DS-ATGO: Dual-Stage Synergistic Learning via Forward Adaptive Threshold and Backward Gradient Optimization for Spiking Neural Networks (1 citation)
- I2E: Real-Time Image-to-Event Conversion for High-Performance Spiking Neural Networks (1 citation)
- Optimization Method for Surrogate Function in Spiking Neural Networks Based on Membrane Potential Distribution (1 citation)
- Parallel Training Time-to-First-Spike Spiking Neural Networks (1 citation)
- SpikCommander: A High-performance Spiking Transformer with Multi-view Learning for Efficient Speech Command Recognition (1 citation)
- Training-Free ANN-to-SNN Conversion for High-Performance Spiking Transformers (1 citation)
- Activation-wise Propagation: A One-Timestep Strategy for Spiking Neural Networks
- Dynamic Weight Adaptation in Spiking Neural Networks Inspired by Biological Homeostasis
and 11 more in the interactive map.
RNNs and normalization · 22 papers
- CloserToMe: A Unified Framework for Accurate and Transferable Latency Prediction Across Heterogeneous Devices (2 citations)
- On Logical Extrapolation for Mazes with Recurrent and Implicit Networks (2 citations)
- Accelerating LLM Inference Throughput via Asynchronous KV Cache Prefetching (1 citation)
- CoRE-Learning with Look-Ahead and Immediate Resource Allocation
- EHL*: Memory-Budgeted Indexing for Ultrafast Optimal Euclidean Pathfinding
- Exploiting Space Folding by Neural Networks
- Feature Integration Spaces: Joint Training Reveals Dual Encoding in Neural Network Representations
- FlashKAT: Understanding and Addressing Performance Bottlenecks in the Kolmogorov-Arnold Transformer
- GlyphShield: Document Watermarking for the Physical World via Vector Typeface Synthesis
- HCF: Hierarchical Cascade Framework for Distributed Multi-Stage Image Compression
- Improving the Accuracy of Dense Retrieval on the Quantized Indexes via Gradient Optimization of the Target Embeddings
- Investigating Data Pruning for Pretraining Biological Foundation Models at Scale
and 10 more in the interactive map.
Mixture of experts · 19 papers
- Sub-MoE: Efficient Mixture-of-Expert LLMs Compression via Subspace Expert Merging (3 citations)
- D2MoRA: Diversity-Regulated Asymmetric MoE-LoRA Decomposition for Efficient Multi-Task Adaptation (2 citations)
- FT-MoE: Sustainable-learning Mixture of Experts for Fault-Tolerant Computing (2 citations)
- CAMERA: Multi-Matrix Joint Compression for MoE Models via Micro-Expert Redundancy Analysis (1 citation)
- CommitMoE: Efficient Fallback-Free MoE Inference with Offloading Under GPU Memory Constraints (1 citation)
- EvoMoE: Expert Evolution in Mixture of Experts for Multimodal Large Language Models (1 citation)
- AdaFuse: Accelerating Dynamic Adapter Inference via Token-Level Pre-Gating and Fused Kernel Optimization
- C-GNN-PRUNE: A Unified Graph-Based Framework for Structure-Aware Pruning of Mixture-of-Experts Models
- CasMoE: A Cascaded Framework for Efficient MoE Inference on Resource-constrained Devices
- Deconstructing Pre-training: Knowledge Attribution Analysis in MoE and Dense Models
- How Many Experts Are Enough? Towards Optimal Semantic Specialization for Mixture-of-Experts
- Hybrid Routing for a Mixture of LoRA Experts
and 7 more in the interactive map.