What Do Self-Supervised Vision Transformers Learn?

We present a comparative study on how and why contrastive learning (CL) and masked image modeling (MIM) differ in their representations and in their performance of downstream tasks. In particular, we demonstrate that self-supervised Vision Transformers (ViTs) have the following properties: (1) CL trains self-attentions to capture longer-range global patterns than MIM, such as the shape of an object, especially in the later layers of the ViT architecture. This CL property helps ViTs linearly separate images in their representation spaces. However, it also makes the self-attentions collapse into homogeneity for all query tokens and heads. Such homogeneity of self-attention reduces the diversity of representations, worsening scalability and dense prediction performance. (2) CL utilizes the low-frequency signals of the representations, but MIM utilizes high-frequencies. Since low- and high-frequency information respectively represent shapes and textures, CL is more shape-oriented and MIM more texture-oriented. (3) CL plays a crucial role in the later layers, while MIM mainly focuses on the early layers. Upon these analyses, we find that CL and MIM can complement each other and observe that even the simplest harmonization can help leverage the advantages of both methods. The code is available at https://github.com/naver-ai/cl-vs-mim.

How Do Vision Transformers Work?

The success of multi-head self-attentions (MSAs) for computer vision is now indisputable. However, little is known about how MSAs work. We present fundamental explanations to help better understand the nature of MSAs. In particular, we demonstrate the following properties of MSAs and Vision Transformers (ViTs): (1) MSAs improve not only accuracy but also generalization by flattening the loss landscapes. Such improvement is primarily attributable to their data specificity, not long-range dependency. On the other hand, ViTs suffer from non-convex losses. Large datasets and loss landscape smoothing methods alleviate this problem; (2) MSAs and Convs exhibit opposite behaviors. For example, MSAs are low-pass filters, but Convs are high-pass filters. Therefore, MSAs and Convs are complementary; (3) Multi-stage neural networks behave like a series connection of small individual models. In addition, MSAs at the end of a stage play a key role in prediction. Based on these insights, we propose AlterNet, a model in which Conv blocks at the end of a stage are replaced with MSA blocks. AlterNet outperforms CNNs not only in large data regimes but also in small data regimes. The code is available at https://github.com/xxxnell/how-do-vits-work.

Blurs Make Results Clearer: Spatial Smoothings to Improve Accuracy, Uncertainty, and Robustness

Bayesian neural networks (BNNs) have shown success in the areas of uncertainty estimation and robustness. However, a crucial challenge prohibits their use in practice: Bayesian NNs require a large number of predictions to produce reliable results, leading to a significant increase in computational cost. To alleviate this issue, we propose spatial smoothing, a method that ensembles neighboring feature map points of CNNs. By simply adding a few blur layers to the models, we empirically show that the spatial smoothing improves accuracy, uncertainty estimation, and robustness of BNNs across a whole range of ensemble sizes. In particular, BNNs incorporating the spatial smoothing achieve high predictive performance merely with a handful of ensembles. Moreover, this method also can be applied to canonical deterministic neural networks to improve the performances. A number of evidences suggest that the improvements can be attributed to the smoothing and flattening of the loss landscape. In addition, we provide a fundamental explanation for prior works - namely, global average pooling, pre-activation, and ReLU6 - by addressing to them as special cases of the spatial smoothing. These not only enhance accuracy, but also improve uncertainty estimation and robustness by making the loss landscape smoother in the same manner as the spatial smoothing. The code is available at https://github.com/xxxnell/spatial-smoothing.

Differentiable Bayesian Neural Network Inference for Data Streams

While deep neural networks (NNs) do not provide the confidence of its prediction, Bayesian neural network (BNN) can estimate the uncertainty of the prediction. However, BNNs have not been widely used in practice due to the computational cost of inference. This prohibitive computational cost is a hindrance especially when processing stream data with low-latency. To address this problem, we propose a novel model which approximate BNNs for data streams. Instead of generating separate prediction for each data sample independently, this model estimates the increments of prediction for a new data sample from the previous predictions. The computational cost of this model is almost the same as that of non-Bayesian NNs. Experiments with semantic segmentation on real-world data show that this model performs significantly faster than BNNs, estimating uncertainty comparable to the results of BNNs.