Generalizing Few-Shot NAS with Gradient Matching

Efficient performance estimation of architectures drawn from large search spaces is essential to Neural Architecture Search. One-Shot methods tackle this challenge by training one supernet to approximate the performance of every architecture in the search space via weight-sharing, thereby drastically reducing the search cost. However, due to coupled optimization between child architectures caused by weight-sharing, One-Shot supernet's performance estimation could be inaccurate, leading to degraded search outcomes. To address this issue, Few-Shot NAS reduces the level of weight-sharing by splitting the One-Shot supernet into multiple separated sub-supernets via edge-wise (layer-wise) exhaustive partitioning. Since each partition of the supernet is not equally important, it necessitates the design of a more effective splitting criterion. In this work, we propose a gradient matching score (GM) that leverages gradient information at the shared weight for making informed splitting decisions. Intuitively, gradients from different child models can be used to identify whether they agree on how to update the shared modules, and subsequently to decide if they should share the same weight. Compared with exhaustive partitioning, the proposed criterion significantly reduces the branching factor per edge. This allows us to split more edges (layers) for a given budget, resulting in substantially improved performance as NAS search spaces usually include dozens of edges (layers). Extensive empirical evaluations of the proposed method on a wide range of search spaces (NASBench-201, DARTS, MobileNet Space), datasets (cifar10, cifar100, ImageNet) and search algorithms (DARTS, SNAS, RSPS, ProxylessNAS, OFA) demonstrate that it significantly outperforms its Few-Shot counterparts while surpassing previous comparable methods in terms of the accuracy of derived architectures.

On Redundancy and Diversity in Cell-based Neural Architecture Search

Searching for the architecture cells is a dominant paradigm in NAS. However, little attention has been devoted to the analysis of the cell-based search spaces even though it is highly important for the continual development of NAS. In this work, we conduct an empirical post-hoc analysis of architectures from the popular cell-based search spaces and find that the existing search spaces contain a high degree of redundancy: the architecture performance is minimally sensitive to changes at large parts of the cells, and universally adopted designs, like the explicit search for a reduction cell, significantly increase the complexities but have very limited impact on the performance. Across architectures found by a diverse set of search strategies, we consistently find that the parts of the cells that do matter for architecture performance often follow similar and simple patterns. By explicitly constraining cells to include these patterns, randomly sampled architectures can match or even outperform the state of the art. These findings cast doubts into our ability to discover truly novel architectures in the existing cell-based search spaces, and inspire our suggestions for improvement to guide future NAS research. Code is available at https://github.com/xingchenwan/cell-based-NAS-analysis.

CODA: A Real-World Road Corner Case Dataset for Object Detection in Autonomous Driving

Contemporary deep-learning object detection methods for autonomous driving usually assume prefixed categories of common traffic participants, such as pedestrians and cars. Most existing detectors are unable to detect uncommon objects and corner cases (e.g., a dog crossing a street), which may lead to severe accidents in some situations, making the timeline for the real-world application of reliable autonomous driving uncertain. One main reason that impedes the development of truly reliably self-driving systems is the lack of public datasets for evaluating the performance of object detectors on corner cases. Hence, we introduce a challenging dataset named CODA that exposes this critical problem of vision-based detectors. The dataset consists of 1500 carefully selected real-world driving scenes, each containing four object-level corner cases (on average), spanning 30+ object categories. On CODA, the performance of standard object detectors trained on large-scale autonomous driving datasets significantly drops to no more than 12.8% in mAR. Moreover, we experiment with the state-of-the-art open-world object detector and find that it also fails to reliably identify the novel objects in CODA, suggesting that a robust perception system for autonomous driving is probably still far from reach. We expect our CODA dataset to facilitate further research in reliable detection for real-world autonomous driving. Our dataset will be released at https://coda-dataset.github.io.

Memory Replay with Data Compression for Continual Learning

Continual learning needs to overcome catastrophic forgetting of the past. Memory replay of representative old training samples has been shown as an effective solution, and achieves the state-of-the-art (SOTA) performance. However, existing work is mainly built on a small memory buffer containing a few original data, which cannot fully characterize the old data distribution. In this work, we propose memory replay with data compression (MRDC) to reduce the storage cost of old training samples and thus increase their amount that can be stored in the memory buffer. Observing that the trade-off between the quality and quantity of compressed data is highly nontrivial for the efficacy of memory replay, we propose a novel method based on determinantal point processes (DPPs) to efficiently determine an appropriate compression quality for currently-arrived training samples. In this way, using a naive data compression algorithm with a properly selected quality can largely boost recent strong baselines by saving more compressed data in a limited storage space. We extensively validate this across several benchmarks of class-incremental learning and in a realistic scenario of object detection for autonomous driving.

Revisiting Over-smoothing in BERT from the Perspective of Graph

Recently over-smoothing phenomenon of Transformer-based models is observed in both vision and language fields. However, no existing work has delved deeper to further investigate the main cause of this phenomenon. In this work, we make the attempt to analyze the over-smoothing problem from the perspective of graph, where such problem was first discovered and explored. Intuitively, the self-attention matrix can be seen as a normalized adjacent matrix of a corresponding graph. Based on the above connection, we provide some theoretical analysis and find that layer normalization plays a key role in the over-smoothing issue of Transformer-based models. Specifically, if the standard deviation of layer normalization is sufficiently large, the output of Transformer stacks will converge to a specific low-rank subspace and result in over-smoothing. To alleviate the over-smoothing problem, we consider hierarchical fusion strategies, which combine the representations from different layers adaptively to make the output more diverse. Extensive experiment results on various data sets illustrate the effect of our fusion method.

DropNAS: Grouped Operation Dropout for Differentiable Architecture Search

Neural architecture search (NAS) has shown encouraging results in automating the architecture design. Recently, DARTS relaxes the search process with a differentiable formulation that leverages weight-sharing and SGD where all candidate operations are trained simultaneously. Our empirical results show that such procedure results in the co-adaption problem and Matthew Effect: operations with fewer parameters would be trained maturely earlier. This causes two problems: firstly, the operations with more parameters may never have the chance to express the desired function since those with less have already done the job; secondly, the system will punish those underperforming operations by lowering their architecture parameter, and they will get smaller loss gradients, which causes the Matthew Effect. In this paper, we systematically study these problems and propose a novel grouped operation dropout algorithm named DropNAS to fix the problems with DARTS. Extensive experiments demonstrate that DropNAS solves the above issues and achieves promising performance. Specifically, DropNAS achieves 2.26% test error on CIFAR-10, 16.39% on CIFAR-100 and 23.4% on ImageNet (with the same training hyperparameters as DARTS for a fair comparison). It is also observed that DropNAS is robust across variants of the DARTS search space. Code is available at https://github.com/wiljohnhong/DropNAS.

Long-tail Recognition via Compositional Knowledge Transfer

In this work, we introduce a novel strategy for long-tail recognition that addresses the tail classes' few-shot problem via training-free knowledge transfer. Our objective is to transfer knowledge acquired from information-rich common classes to semantically similar, and yet data-hungry, rare classes in order to obtain stronger tail class representations. We leverage the fact that class prototypes and learned cosine classifiers provide two different, complementary representations of class cluster centres in feature space, and use an attention mechanism to select and recompose learned classifier features from common classes to obtain higher quality rare class representations. Our knowledge transfer process is training free, reducing overfitting risks, and can afford continual extension of classifiers to new classes. Experiments show that our approach can achieve significant performance boosts on rare classes while maintaining robust common class performance, outperforming directly comparable state-of-the-art models.

Layer-Parallel Training of Residual Networks with Auxiliary-Variable Networks

Gradient-based methods for the distributed training of residual networks (ResNets) typically require a forward pass of the input data, followed by back-propagating the error gradient to update model parameters, which becomes time-consuming as the network goes deeper. To break the algorithmic locking and exploit synchronous module parallelism in both the forward and backward modes, auxiliary-variable methods have attracted much interest lately but suffer from significant communication overhead and lack of data augmentation. In this work, a novel joint learning framework for training realistic ResNets across multiple compute devices is established by trading off the storage and recomputation of external auxiliary variables. More specifically, the input data of each independent processor is generated from its low-capacity auxiliary network (AuxNet), which permits the use of data augmentation and realizes forward unlocking. The backward passes are then executed in parallel, each with a local loss function that originates from the penalty or augmented Lagrangian (AL) methods. Finally, the proposed AuxNet is employed to reproduce the updated auxiliary variables through an end-to-end training process. We demonstrate the effectiveness of our methods on ResNets and WideResNets across CIFAR-10, CIFAR-100, and ImageNet datasets, achieving speedup over the traditional layer-serial training method while maintaining comparable testing accuracy.

Understanding Square Loss in Training Overparametrized Neural Network Classifiers

Deep learning has achieved many breakthroughs in modern classification tasks. Numerous architectures have been proposed for different data structures but when it comes to the loss function, the cross-entropy loss is the predominant choice. Recently, several alternative losses have seen revived interests for deep classifiers. In particular, empirical evidence seems to promote square loss but a theoretical justification is still lacking. In this work, we contribute to the theoretical understanding of square loss in classification by systematically investigating how it performs for overparametrized neural networks in the neural tangent kernel (NTK) regime. Interesting properties regarding the generalization error, robustness, and calibration error are revealed. We consider two cases, according to whether classes are separable or not. In the general non-separable case, fast convergence rate is established for both misclassification rate and calibration error. When classes are separable, the misclassification rate improves to be exponentially fast. Further, the resulting margin is proven to be lower bounded away from zero, providing theoretical guarantees for robustness. We expect our findings to hold beyond the NTK regime and translate to practical settings. To this end, we conduct extensive empirical studies on practical neural networks, demonstrating the effectiveness of square loss in both synthetic low-dimensional data and real image data. Comparing to cross-entropy, square loss has comparable generalization error but noticeable advantages in robustness and model calibration.