重要的是要了解流行的正则化方法如何帮助神经网络训练找到良好的概括解决方案。在这项工作中，我们从理论上得出了辍学的隐式正则化，并研究了损失函数的Hessian矩阵与辍学噪声的协方差矩阵之间的关系，并由一系列实验支持。然后，我们在数值上研究了辍学的隐式正规化的两个含义，这直觉上合理化了辍学有助于概括。首先，我们发现辍学的训练与实验中的标准梯度下降训练相比，发现具有最低最小的神经网络，而隐式正则化是找到平坦溶液的关键。其次，经过辍学的训练，隐藏神经元的输入权重（隐藏神经元的输入权重由其输入层到隐藏的神经元及其偏见项组成），往往会凝结在孤立的方向上。凝结是非线性学习过程中的一个功能，它使神经网络的复杂性低。尽管我们的理论主要集中在最后一个隐藏层中使用的辍学，但我们的实验适用于训练神经网络中的一般辍学。这项工作指出了与随机梯度下降相比，辍学的独特特征，是完全理解辍学的重要基础。

我们证明了深度神经网络（NNS）的损失景观的一般嵌入原理，其解除了NNS的损失景观的层次结构，即NN的损失景观包含所有较窄NN的所有关键点。通过构建一类临界嵌入来获得该结果，该临界嵌入物将较窄的Nn的任何临界点映射到具有相同输出功能的目标Nn的临界点。通过发现广泛的一般兼容性嵌入式，我们提供了嵌入来自NNS的关键点的关键子多种尺寸的总估计。我们进一步证明了任何临界嵌入的Irfreversiblility属性，即临界点的Hessian矩阵的负/零/正小叶值的数量可能增加，但由于NN通过嵌入越来越宽，因此从未减少。使用一般兼容的临界嵌入的特殊实现，我们证明了一个严格的必要条件，以便是一个完全不变的临界点，从未成为任何关键嵌入的严格鞍端。该结果暗示宽NNS中严格鞍点的常见，这可能是在实践中广泛观察到的宽NNS易于优化的重要原因。

虽然辍学措施在深度学习中取得了巨大成功，但对于如何帮助训练在高维参数空间中发现良好的普遍化解决方案，众所周知。在这项工作中，我们表明，与标准梯度血统训练相比，辍学的训练发现了一个更平坦的最小值。我们进一步研究了通过实验辍学发现更平坦的最小值的潜在机制。我们提出了一种{\ IT方差原理}，噪声横向的较小方向噪声的变化更大。现有的作品表明，SGD满足方差原则，这导致趋势趋势达到更少的最小值。我们的工作表明，辍学引起的噪音也满足了差异原则，解释了为什么辍学发现更漂亮的最小值。一般而言，我们的工作指出，方差原则是辍学和SGD之间的重要相似性，导致培训寻找更平坦的最小值并获得良好的概括。

了解深神经网络的损失景观结构（DNN）显然是重要的。在这项工作中，我们证明了一个嵌入原则，即DNN“包含”所有较窄DNN的所有关键点的损失景观。更确切地说，我们提出了一个临界嵌入，使得任何临界点，例如较窄的DNN的临界点，例如局部或全局最小值，可以嵌入到目标DNN的临界点/超平面，具有更高的退化性并保持DNN输出功能。关键点的嵌入结构与损耗功能和训练数据无关，显示与蛋白质折叠等其他非凸起问题的显着差异。凭经验，我们发现宽DNN通常被嵌入来自窄DNN的高度简并关键点引起。嵌入原理为广泛DNN的普遍易于优化提供了解释，并且在训练期间揭开潜在的隐式低复杂性正则化。总体而言，我们的工作为DNNS的损失景观提供了骨架及其含义，可以在附近预期更精确和全面的理解

Masked image modeling (MIM) performs strongly in pre-training large vision Transformers (ViTs). However, small models that are critical for real-world applications cannot or only marginally benefit from this pre-training approach. In this paper, we explore distillation techniques to transfer the success of large MIM-based pre-trained models to smaller ones. We systematically study different options in the distillation framework, including distilling targets, losses, input, network regularization, sequential distillation, etc, revealing that: 1) Distilling token relations is more effective than CLS token- and feature-based distillation; 2) An intermediate layer of the teacher network as target perform better than that using the last layer when the depth of the student mismatches that of the teacher; 3) Weak regularization is preferred; etc. With these findings, we achieve significant fine-tuning accuracy improvements over the scratch MIM pre-training on ImageNet-1K classification, using all the ViT-Tiny, ViT-Small, and ViT-base models, with +4.2%/+2.4%/+1.4% gains, respectively. Our TinyMIM model of base size achieves 52.2 mIoU in AE20K semantic segmentation, which is +4.1 higher than the MAE baseline. Our TinyMIM model of tiny size achieves 79.6% top-1 accuracy on ImageNet-1K image classification, which sets a new record for small vision models of the same size and computation budget. This strong performance suggests an alternative way for developing small vision Transformer models, that is, by exploring better training methods rather than introducing inductive biases into architectures as in most previous works. Code is available at https://github.com/OliverRensu/TinyMIM.

In this paper, we propose a robust 3D detector, named Cross Modal Transformer (CMT), for end-to-end 3D multi-modal detection. Without explicit view transformation, CMT takes the image and point clouds tokens as inputs and directly outputs accurate 3D bounding boxes. The spatial alignment of multi-modal tokens is performed implicitly, by encoding the 3D points into multi-modal features. The core design of CMT is quite simple while its performance is impressive. CMT obtains 73.0% NDS on nuScenes benchmark. Moreover, CMT has a strong robustness even if the LiDAR is missing. Code will be released at https://github.com/junjie18/CMT.

Dataset distillation has emerged as a prominent technique to improve data efficiency when training machine learning models. It encapsulates the knowledge from a large dataset into a smaller synthetic dataset. A model trained on this smaller distilled dataset can attain comparable performance to a model trained on the original training dataset. However, the existing dataset distillation techniques mainly aim at achieving the best trade-off between resource usage efficiency and model utility. The security risks stemming from them have not been explored. This study performs the first backdoor attack against the models trained on the data distilled by dataset distillation models in the image domain. Concretely, we inject triggers into the synthetic data during the distillation procedure rather than during the model training stage, where all previous attacks are performed. We propose two types of backdoor attacks, namely NAIVEATTACK and DOORPING. NAIVEATTACK simply adds triggers to the raw data at the initial distillation phase, while DOORPING iteratively updates the triggers during the entire distillation procedure. We conduct extensive evaluations on multiple datasets, architectures, and dataset distillation techniques. Empirical evaluation shows that NAIVEATTACK achieves decent attack success rate (ASR) scores in some cases, while DOORPING reaches higher ASR scores (close to 1.0) in all cases. Furthermore, we conduct a comprehensive ablation study to analyze the factors that may affect the attack performance. Finally, we evaluate multiple defense mechanisms against our backdoor attacks and show that our attacks can practically circumvent these defense mechanisms.

Blind image quality assessment (BIQA) remains challenging due to the diversity of distortion and image content variation, which complicate the distortion patterns crossing different scales and aggravate the difficulty of the regression problem for BIQA. However, existing BIQA methods often fail to consider multi-scale distortion patterns and image content, and little research has been done on learning strategies to make the regression model produce better performance. In this paper, we propose a simple yet effective Progressive Multi-Task Image Quality Assessment (PMT-IQA) model, which contains a multi-scale feature extraction module (MS) and a progressive multi-task learning module (PMT), to help the model learn complex distortion patterns and better optimize the regression issue to align with the law of human learning process from easy to hard. To verify the effectiveness of the proposed PMT-IQA model, we conduct experiments on four widely used public datasets, and the experimental results indicate that the performance of PMT-IQA is superior to the comparison approaches, and both MS and PMT modules improve the model's performance.

Automatic music generation with artificial intelligence typically requires a large amount of data which is hard to obtain for many less common genres and musical instruments. To tackle this issue, we present ongoing work and preliminary findings on the possibility for deep models to transfer knowledge from language to music, by finetuning large language models pre-trained on a massive text corpus on only hundreds of MIDI files of drum performances. We show that by doing so, one of the largest, state-of-the-art models (GPT3) is capable of generating reasonable drum grooves, while models that are not pre-trained (Transformer) shows no such ability beyond naive repetition. Evaluating generated music is a challenging task, more so is evaluating drum grooves with little precedence in literature. Hence, we propose a tailored structural evaluation method and analyze drum grooves produced by GPT3 compared to those played by human professionals, exposing the strengths and weaknesses of such generation by language-to-music transfer. Our findings suggest that language-to-music transfer learning with large language models is viable and promising.

Few Shot Instance Segmentation (FSIS) requires models to detect and segment novel classes with limited several support examples. In this work, we explore a simple yet unified solution for FSIS as well as its incremental variants, and introduce a new framework named Reference Twice (RefT) to fully explore the relationship between support/query features based on a Transformer-like framework. Our key insights are two folds: Firstly, with the aid of support masks, we can generate dynamic class centers more appropriately to re-weight query features. Secondly, we find that support object queries have already encoded key factors after base training. In this way, the query features can be enhanced twice from two aspects, i.e., feature-level and instance-level. In particular, we firstly design a mask-based dynamic weighting module to enhance support features and then propose to link object queries for better calibration via cross-attention. After the above steps, the novel classes can be improved significantly over our strong baseline. Additionally, our new framework can be easily extended to incremental FSIS with minor modification. When benchmarking results on the COCO dataset for FSIS, gFSIS, and iFSIS settings, our method achieves a competitive performance compared to existing approaches across different shots, e.g., we boost nAP by noticeable +8.2/+9.4 over the current state-of-the-art FSIS method for 10/30-shot. We further demonstrate the superiority of our approach on Few Shot Object Detection. Code and model will be available.