The success of deep learning is inseparable from normalization layers. Researchers have proposed various normalization functions, and each of them has both advantages and disadvantages. In response, efforts have been made to design a unified normalization function that combines all normalization procedures and mitigates their weaknesses. We also proposed a new normalization function called Adaptive Fusion Normalization. Through experiments, we demonstrate AFN outperforms the previous normalization techniques in domain generalization and image classification tasks.

A simple design recipe for deep Transformers is to compose identical building blocks. But standard transformer blocks are far from simple, interweaving attention and MLP sub-blocks with skip connections & normalisation layers in precise arrangements. This complexity leads to brittle architectures, where seemingly minor changes can significantly reduce training speed, or render models untrainable. In this work, we ask to what extent the standard transformer block can be simplified? Combining signal propagation theory and empirical observations, we motivate modifications that allow many block components to be removed with no loss of training speed, including skip connections, projection or value parameters, sequential sub-blocks and normalisation layers. In experiments on both autoregressive decoder-only and BERT encoder-only models, our simplified transformers emulate the per-update training speed and performance of standard transformers, while enjoying 15% faster training throughput, and using 15% fewer parameters. The transformer architecture (Vaswani et al., 2017) is arguably the workhorse behind many recent successes in deep learning. A simple way to construct a deep transformer architecture is by stacking multiple identical transformer "blocks" one after another in sequence. Each block, however, is more complicated and consists of many different components, which need to be combined in specific arrangements in order to achieve good performance. Surprisingly, the base transformer block has changed very little since its inception, despite attracting the interest of many researchers. In this work, we study whether the standard transformer block can be simplified. More specifically, we probe the necessity of several block components, including skip connections, projection/value matrices, sequential sub-blocks and normalisation layers.

The linearised Laplace method for estimating model uncertainty has received renewed attention in the Bayesian deep learning community. The method provides reliable error bars and admits a closed-form expression for the model evidence, allowing for scalable selection of model hyperparameters. In this work, we examine the assumptions behind this method, particularly in conjunction with model selection. We show that these interact poorly with some now-standard tools of deep learning--stochastic approximation methods and normalisation layers--and make recommendations for how to better adapt this classic method to the modern setting. We provide theoretical support for our recommendations and validate them empirically on MLPs, classic CNNs, residual networks with and without normalisation layers, generative autoencoders and transformers.

Existing person re-identification (Re-ID) methods mostly follow a centralised learning paradigm which shares all training data to a collection for model learning. This paradigm is limited when data from different sources cannot be shared due to privacy concerns. To resolve this problem, two recent works have introduced decentralised (federated) Re-ID learning for constructing a globally generalised model (server)without any direct access to local training data nor shared data across different source domains (clients). However, these methods are poor on how to adapt the generalised model to maximise its performance on individual client domain Re-ID tasks having different Re-ID label spaces, due to a lack of understanding of data heterogeneity across domains. We call this poor 'model personalisation'. In this work, we present a new Selective Knowledge Aggregation approach to decentralised person Re-ID to optimise the trade-off between model personalisation and generalisation. Specifically, we incorporate attentive normalisation into the normalisation layers in a deep ReID model and propose to learn local normalisation layers specific to each domain, which are decoupled from the global model aggregation in federated Re-ID learning. This helps to preserve model personalisation knowledge on each local client domain and learn instance-specific information. Further, we introduce a dual local normalisation mechanism to learn generalised normalisation layers in each local model, which are then transmitted to the global model for central aggregation. This facilitates selective knowledge aggregation on the server to construct a global generalised model for out-of-the-box deployment on unseen novel domains. Extensive experiments on eight person Re-ID datasets show that the proposed approach to decentralised Re-ID significantly outperforms the state-of-the-art decentralised methods.