In recent years, Vision Transformers have attracted increasing interest from computer vision researchers. However, the advantage of these transformers over CNNs is only fully manifested when trained over a large dataset, mainly due to the reduced inductive bias towards spatial locality within the transformer's self-attention mechanism. In this work, we present a data-efficient vision transformer that does not rely on self-attention. Instead, it employs a novel generalization to multiple axes of the very recent Hyena layer. We propose several alternative approaches for obtaining this generalization and delve into their unique distinctions and considerations from both empirical and theoretical perspectives. Our empirical findings indicate that the proposed Hyena N-D layer boosts the performance of various Vision Transformer architectures, such as ViT, Swin, and DeiT across multiple datasets. Furthermore, in the small dataset regime, our Hyena-based ViT is favorable to ViT variants from the recent literature that are specifically designed for solving the same challenge, i.e., working with small datasets or incorporating image-specific inductive bias into the self-attention mechanism. Finally, we show that a hybrid approach that is based on Hyena N-D for the first layers in ViT, followed by layers that incorporate conventional attention, consistently boosts the performance of various vision transformer architectures.

Generative modeling of 3D LiDAR data is an emerging task with promising applications for autonomous mobile robots, such as scalable simulation, scene manipulation, and sparse-to-dense completion of LiDAR point clouds. Existing approaches have shown the feasibility of image-based LiDAR data generation using deep generative models while still struggling with the fidelity of generated data and training instability. In this work, we present R2DM, a novel generative model for LiDAR data that can generate diverse and high-fidelity 3D scene point clouds based on the image representation of range and reflectance intensity. Our method is based on the denoising diffusion probabilistic models (DDPMs), which have demonstrated impressive results among generative model frameworks and have been significantly progressing in recent years. To effectively train DDPMs on the LiDAR domain, we first conduct an in-depth analysis regarding data representation, training objective, and spatial inductive bias. Based on our designed model R2DM, we also introduce a flexible LiDAR completion pipeline using the powerful properties of DDPMs. We demonstrate that our method outperforms the baselines on the generation task of KITTI-360 and KITTI-Raw datasets and the upsampling task of KITTI-360 datasets. Our code and pre-trained weights will be available at https://github.com/kazuto1011/r2dm.

Though deep learning models have taken on commercial and political relevance, many aspects of their training and operation remain poorly understood. This has sparked interest in "science of deep learning" projects, many of which are run at scale and require enormous amounts of time, money, and electricity. But how much of this research really needs to occur at scale? In this paper, we introduce MNIST-1D: a minimalist, low-memory, and low-compute alternative to classic deep learning benchmarks. The training examples are 20 times smaller than MNIST examples yet they differentiate more clearly between linear, nonlinear, and convolutional models which attain 32, 68, and 94% accuracy respectively (these models obtain 94, 99+, and 99+% on MNIST). Then we present example use cases which include measuring the spatial inductive biases of lottery tickets, observing deep double descent, and metalearning an activation function.