The field of artificial intelligence is moving towards more efficient and robust transformer architectures and diffusion models. Recent research has focused on improving the performance and stability of these models, with a particular emphasis on understanding how they process and represent information. One key direction is the development of more effective methods for updating and fine-tuning transformer models, such as the use of rank-1 patches and frequency-energy constrained routing. Another important area of research is the development of more robust and reliable diffusion models, including the use of latent diffusion inversion and entropy-guided prioritized progressive learning. Noteworthy papers in this area include 'Equivalence of Context and Parameter Updates in Modern Transformer Blocks', which provides a general framework for understanding how transformer models transmute prompts into effective weights, and 'FeRA: Frequency-Energy Constrained Routing for Effective Diffusion Adaptation Fine-Tuning', which proposes a frequency driven fine tuning framework for diffusion models. Overall, these advancements have the potential to improve the performance and efficiency of a wide range of AI applications, from natural language processing to computer vision.
Advancements in Transformer Architecture and Diffusion Models
Sources
Equivalence of Context and Parameter Updates in Modern Transformer Blocks
FeRA: Frequency-Energy Constrained Routing for Effective Diffusion Adaptation Fine-Tuning
Subtract the Corruption: Training-Data-Free Corrective Machine Unlearning using Task Arithmetic
Deterministic Continuous Replacement: Fast and Stable Module Replacement in Pretrained Transformers
Evaluating Dataset Watermarking for Fine-tuning Traceability of Customized Diffusion Models: A Comprehensive Benchmark and Removal Approach
SPQR: A Standardized Benchmark for Modern Safety Alignment Methods in Text-to-Image Diffusion Models
ModHiFi: Identifying High Fidelity predictive components for Model Modification
Learning to Clean: Reinforcement Learning for Noisy Label Correction
Latent Diffusion Inversion Requires Understanding the Latent Space
Post-Pruning Accuracy Recovery via Data-Free Knowledge Distillation
Pre-train to Gain: Robust Learning Without Clean Labels
Operationalizing Quantized Disentanglement
Which Layer Causes Distribution Deviation? Entropy-Guided Adaptive Pruning for Diffusion and Flow Models
Efficient Training for Human Video Generation with Entropy-Guided Prioritized Progressive Learning
AuthenLoRA: Entangling Stylization with Imperceptible Watermarks for Copyright-Secure LoRA Adapters
Data Exfiltration by Compression Attack: Definition and Evaluation on Medical Image Data
Illuminating the Black Box: Real-Time Monitoring of Backdoor Unlearning in CNNs via Explainable AI
Controlling changes to attention logits