LampQ: Towards Accurate Layer-wise Mixed Precision Quantization for Vision Transformers

Minimizing inconsistencies in AI systems is crucial for reducing overall error rates. In image classification, negative flips occur when updated models misclassify previously correctly classified samples. This issue intensifies with the addition of new training classes, which can reduce the margin between classes and introduce conflicting patterns. To address this, a novel approach is proposed that preserves the original model's margins while improving performance, utilizing a margin-calibration term to enhance class separation.

The article presents SemanticNN, a novel semantic codec designed for extremely weak embedded devices in the Internet of Things (IoT). It addresses the challenges of integrating artificial intelligence (AI) on such devices, which often face resource limitations and unreliable network conditions. SemanticNN focuses on achieving semantic-level correctness despite bit-level errors, utilizing a Bit Error Rate (BER)-aware decoder and a Soft Quantization (SQ)-based encoder to enhance collaborative inference offloading.

Artificial intelligence (AI) in media has seen rapid advancements over the past decade, particularly with the introduction of Generative Adversarial Networks (GANs) and diffusion models, which have enhanced photorealistic image generation. However, these developments have also led to challenges in distinguishing between real and synthetic content, as evidenced by the rise of deepfakes. Many detection models utilizing deep learning methods like Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have been created, but they often struggle with generalization and multimodal data.

The paper introduces Synthetic Object Compositions (SOC), a novel data synthesis pipeline aimed at enhancing computer vision tasks such as instance segmentation, visual grounding, and object detection. SOC addresses the limitations of traditional datasets, which are often costly and biased, by generating high-quality synthetic object segments through advanced techniques like 3D geometric layout augmentation. This approach promises improved accuracy and diversity in visual data, essential for applications ranging from robotic perception to photo editing.

Image deblurring is a crucial aspect of computer vision, focused on restoring sharp images from blurry ones caused by motion or camera shake. Traditional deep learning methods, including CNNs and Vision Transformers (ViTs), face challenges with complex blurs and high computational demands. A new dual-domain architecture integrates Vision Transformers with a frequency-domain FFT-ReLU module, enhancing the ability to suppress blur artifacts while preserving details, achieving superior performance metrics such as PSNR and SSIM in extensive experiments.

The paper titled 'Convergence Bound and Critical Batch Size of Muon Optimizer' presents a theoretical analysis of the Muon optimizer, which has shown strong empirical performance and is proposed as a successor to AdamW. The study provides convergence proofs for Muon across four practical settings, examining its behavior with and without Nesterov momentum and weight decay. It highlights that the inclusion of weight decay results in tighter theoretical bounds and identifies the critical batch size that minimizes training costs, validated through experiments in image classification and language modeling.