Vision transformers (ViTs) have pushed the state-of-the-art for various visual recognition tasks by patch-wise image tokenization followed by self-attention. However, the employment of self-attention modules results in a quadratic complexity in both computation and memory usage. Various attempts on approximating the self-attention computation with linear complexity have been made in Natural Language Processing. However, an in-depth analysis in this work shows that they are either theoretically flawed or empirically ineffective for visual recognition. We further identify that their limitations are rooted in keeping the softmax self-attention during approximations. Specifically, conventional self-attention is computed by normalizing the scaled dot-product between token feature vectors. Keeping this softmax operation challenges any subsequent linearization efforts. Based on this insight, for the first time, a softmax-free transformer or SOFT is proposed. To remove softmax in self-attention, Gaussian kernel function is used to replace the dot-product similarity without further normalization. This enables a full self-attention matrix to be approximated via a low-rank matrix decomposition. The robustness of the approximation is achieved by calculating its Moore-Penrose inverse using a Newton-Raphson method. Extensive experiments on ImageNet show that our SOFT significantly improves the computational efficiency of existing ViT variants. Crucially, with a linear complexity, much longer token sequences are permitted in SOFT, resulting in superior trade-off between accuracy and complexity.
iALS is a popular algorithm for learning matrix factorization models from implicit feedback with alternating least squares. This algorithm was invented over a decade ago but still shows competitive quality compared to recent approaches like VAE, EASE, SLIM, or NCF. Due to a computational trick that avoids negative sampling, iALS is very efficient especially for large item catalogues. However, iALS does not scale well with large embedding dimensions, d, due to its cubic runtime dependency on d. Coordinate descent variations, iCD, have been proposed to lower the complexity to quadratic in d. In this work, we show that iCD approaches are not well suited for modern processors and can be an order of magnitude slower than a careful iALS implementation for small to mid scale embedding sizes (d ~ 100) and only perform better than iALS on large embeddings d ~ 1000. We propose a new solver iALS++ that combines the advantages of iALS in terms of vector processing with a low computational complexity as in iCD. iALS++ is an order of magnitude faster than iCD both for small and large embedding dimensions. It can solve benchmark problems like Movielens 20M or Million Song Dataset even for 1000 dimensional embedding vectors in a few minutes.
Matrix factorization learned by implicit alternating least squares (iALS) is a popular baseline in recommender system research publications. iALS is known to be one of the most computationally efficient and scalable collaborative filtering methods. However, recent studies suggest that its prediction quality is not competitive with the current state of the art, in particular autoencoders and other item-based collaborative filtering methods. In this work, we revisit the iALS algorithm and present a bag of tricks that we found useful when applying iALS. We revisit four well-studied benchmarks where iALS was reported to perform poorly and show that with proper tuning, iALS is highly competitive and outperforms any method on at least half of the comparisons. We hope that these high quality results together with iALS's known scalability spark new interest in applying and further improving this decade old technique.
Text-based person search (TBPS) aims at retrieving a target person from an image gallery with a descriptive text query. Solving such a fine-grained cross-modal retrieval task is challenging, which is further hampered by the lack of large-scale datasets. In this paper, we present a framework with two novel components to handle the problems brought by limited data. Firstly, to fully utilize the existing small-scale benchmarking datasets for more discriminative feature learning, we introduce a cross-modal momentum contrastive learning framework to enrich the training data for a given mini-batch. Secondly, we propose to transfer knowledge learned from existing coarse-grained large-scale datasets containing image-text pairs from drastically different problem domains to compensate for the lack of TBPS training data. A transfer learning method is designed so that useful information can be transferred despite the large domain gap. Armed with these components, our method achieves new state of the art on the CUHK-PEDES dataset with significant improvements over the prior art in terms of Rank-1 and mAP. Our code is available at https://github.com/BrandonHanx/TextReID.
The use of channel-wise attention in CNN based speaker representation networks has achieved remarkable performance in speaker verification (SV). But these approaches do simple averaging on time and frequency feature maps before channel-wise attention learning and ignore the essential mutual interaction among temporal, channel as well as frequency scales. To address this problem, we propose the Duality Temporal-Channel-Frequency (DTCF) attention to re-calibrate the channel-wise features with aggregation of global context on temporal and frequency dimensions. Specifically, the duality attention - time-channel (T-C) attention as well as frequency-channel (F-C) attention - aims to focus on salient regions along the T-C and F-C feature maps that may have more considerable impact on the global context, leading to more discriminative speaker representations. We evaluate the effectiveness of the proposed DTCF attention on the CN-Celeb and VoxCeleb datasets. On the CN-Celeb evaluation set, the EER/minDCF of ResNet34-DTCF are reduced by 0.63%/0.0718 compared with those of ResNet34-SE. On VoxCeleb1-O, VoxCeleb1-E and VoxCeleb1-H evaluation sets, the EER/minDCF of ResNet34-DTCF achieve 0.36%/0.0263, 0.39%/0.0382 and 0.74%/0.0753 reductions compared with those of ResNet34-SE.
Electroencephalogram (EEG) can objectively reflect emotional state and changes. However, the transmission mechanism of EEG in the brain and its internal relationship with emotion are still ambiguous to human beings. This paper presents a novel approach to EEG emotion recognition built exclusively on self-attention over the spectrum, space, and time dimensions to explore the contribution of different EEG electrodes and temporal slices to specific emotional states. Our method, named EEG emotion Transformer (EeT), adapts the conventional Transformer architecture to EEG signals by enabling spatiospectral feature learning directly from the sequences of EEG signals. Our experimental results demonstrate that "joint attention" where temporal and spatial attention are applied simultaneously within each block, leads to the best emotion recognition accuracy among the design choices. In addition, compared with other competitive methods, the proposed method achieves state-of-art results on SEED and SEED-IV datasets.
This paper proposes an unsupervised cross-modality domain adaptation approach based on pixel alignment and self-training. Pixel alignment transfers ceT1 scans to hrT2 modality, helping to reduce domain shift in the training segmentation model. Self-training adapts the decision boundary of the segmentation network to fit the distribution of hrT2 scans. Experiment results show that PAST has outperformed the non-UDA baseline significantly, and it received rank-2 on CrossMoDA validation phase Leaderboard with a mean Dice score of 0.8395.
In this report, we describe the Beijing ZKJ-NPU team submission to the VoxCeleb Speaker Recognition Challenge 2021 (VoxSRC-21). We participated in the fully supervised speaker verification track 1 and track 2. In the challenge, we explored various kinds of advanced neural network structures with different pooling layers and objective loss functions. In addition, we introduced the ResNet-DTCF, CoAtNet and PyConv networks to advance the performance of CNN-based speaker embedding model. Moreover, we applied embedding normalization and score normalization at the evaluation stage. By fusing 11 and 14 systems, our final best performances (minDCF/EER) on the evaluation trails are 0.1205/2.8160% and 0.1175/2.8400% respectively for track 1 and 2. With our submission, we came to the second place in the challenge for both tracks.
In this paper, we conduct wireless channel measurements in indoor corridor scenarios at 2.4, 5 and 6 GHz bands with bandwidth of 320 MHz. The measurement results of channel characteristics at different frequency bands such as average power delay profile (APDP), path loss (PL), delay spread (DS), and Ricean K factor (KF) are presented and analyzed. It is found that the PL exponent (PLE) and PL offset \beta in the floating-intercept (FI) model tend to increase with the increase of frequency. The DS and KF values of the three frequency bands in line of sight (LOS) scenario are basically the same. These results are significant for the design of communication systems.