Neighborhood attention reduces the cost of self attention by restricting each token's attention span to its nearest neighbors. This restriction, parameterized by a window size and dilation factor, draws a spectrum of possible attention patterns between linear projection and self attention. Neighborhood attention, and more generally sliding window attention patterns, have long been bounded by infrastructure, particularly in higher-rank spaces (2-D and 3-D), calling for the development of custom kernels, which have been limited in either functionality, or performance, if not both. In this work, we aim to massively improve upon existing infrastructure by providing two new methods for implementing neighborhood attention. We first show that neighborhood attention can be represented as a batched GEMM problem, similar to standard attention, and implement it for 1-D and 2-D neighborhood attention.

Neighborhood attention reduces the cost of self attention by restricting each token's attention span to its nearest neighbors. This restriction, parameterized by a window size and dilation factor, draws a spectrum of possible attention patterns between linear projection and self attention. Neighborhood attention, and more generally sliding window attention patterns, have long been bounded by infrastructure, particularly in higher-rank spaces (2-D and 3-D), calling for the development of custom kernels, which have been limited in either functionality, or performance, if not both.

Neighborhood attention reduces the cost of self attention by restricting each token's attention span to its nearest neighbors. This restriction, parameterized by a window size and dilation factor, draws a spectrum of possible attention patterns between linear projection and self attention. Neighborhood attention, and more generally sliding window attention patterns, have long been bounded by infrastructure, particularly in higher-rank spaces (2-D and 3-D), calling for the development of custom kernels, which have been limited in either functionality, or performance, if not both. In this work, we aim to massively improve upon existing infrastructure by providing two new methods for implementing neighborhood attention. We first show that neighborhood attention can be represented as a batched GEMM problem, similar to standard attention, and implement it for 1-D and 2-D neighborhood attention.

Many sparse attention mechanisms such as Neighborhood Attention have typically failed to consistently deliver speedup over the self attention baseline. This is largely due to the level of complexity in attention infrastructure, and the rapid evolution of AI hardware architecture. At the same time, many state-of-the-art foundational models, particularly in computer vision, are heavily bound by attention, and need reliable sparsity to escape the O(n^2) complexity. In this paper, we study a class of promising sparse attention mechanisms that focus on locality, and aim to develop a better analytical model of their performance improvements. We first introduce Generalized Neighborhood Attention (GNA), which can describe sliding window, strided sliding window, and blocked attention. We then consider possible design choices in implementing these approaches, and create a simulator that can provide much more realistic speedup upper bounds for any given setting. Finally, we implement GNA on top of a state-of-the-art fused multi-headed attention (FMHA) kernel designed for the NVIDIA Blackwell architecture in CUTLASS. Our implementation can fully realize the maximum speedup theoretically possible in many perfectly block-sparse cases, and achieves an effective utilization of 1.3 petaFLOPs/second in FP16. In addition, we plug various GNA configurations into off-the-shelf generative models, such as Cosmos-7B, HunyuanVideo, and FLUX, and show that it can deliver 28% to 46% end-to-end speedup on B200 without any fine-tuning. We will open source our simulator and Blackwell kernels directly through the NATTEN project.

The literature is abundant with methodologies focusing on using transformer architectures due to their prominence in wireless signal processing and their capability to capture long-range dependencies via attention mechanisms. In particular, depthwise separable convolutions enhance parameter efficiency for the process of high-dimensional data characteristics of MIMO systems. In this work, we introduce a novel unsupervised deep learning framework that integrates depthwise separable convolutions and transformers to generate beamforming weights under imperfect channel state information (CSI) for a multi-user single-input multiple-output (MU-SIMO) system in dense urban environments. The primary goal is to enhance throughput by maximizing sum-rate while ensuring reliable communication. Spectral efficiency and block error rate (BLER) are considered as performance metrics. Experiments are carried out under various conditions to compare the performance of the proposed NNBF framework against baseline methods zero-forcing beamforming (ZFBF) and minimum mean square error (MMSE) beamforming. Experimental results demonstrate the superiority of the proposed framework over the baseline techniques.

Open-ended story visualization is a challenging task that involves generating coherent image sequences from a given storyline. One of the main difficulties is maintaining character consistency while creating natural and contextually fitting scenes--an area where many existing methods struggle. In this paper, we propose an enhanced Transformer module that uses separate self attention and cross attention mechanisms, leveraging prior knowledge from pre-trained diffusion models to ensure logical scene creation. The isolated self attention mechanism improves character consistency by refining attention maps to reduce focus on irrelevant areas and highlight key features of the same character. Meanwhile, the isolated cross attention mechanism independently processes each character's features, avoiding feature fusion and further strengthening consistency. Notably, our method is training-free, allowing the continuous generation of new characters and storylines without re-tuning. Both qualitative and quantitative evaluations show that our approach outperforms current methods, demonstrating its effectiveness.

Do the rich representations of multi-modal diffusion transformers (DiTs) exhibit unique properties that enhance their interpretability? We introduce ConceptAttention, a novel method that leverages the expressive power of DiT attention layers to generate high-quality saliency maps that precisely locate textual concepts within images. Without requiring additional training, ConceptAttention repurposes the parameters of DiT attention layers to produce highly contextualized concept embeddings, contributing the major discovery that performing linear projections in the output space of DiT attention layers yields significantly sharper saliency maps compared to commonly used cross-attention mechanisms. Remarkably, ConceptAttention even achieves state-of-the-art performance on zero-shot image segmentation benchmarks, outperforming 11 other zero-shot interpretability methods on the ImageNet-Segmentation dataset and on a single-class subset of PascalVOC. Our work contributes the first evidence that the representations of multi-modal DiT models like Flux are highly transferable to vision tasks like segmentation, even outperforming multi-modal foundation models like CLIP.

The blood-brain barrier (BBB) serves as a protective barrier that separates the brain from the circulatory system, regulating the passage of substances into the central nervous system. Assessing the BBB permeability of potential drugs is crucial for effective drug targeting. However, traditional experimental methods for measuring BBB permeability are challenging and impractical for large-scale screening. Consequently, there is a need to develop computational approaches to predict BBB permeability. This paper proposes a GPS Transformer architecture augmented with Self Attention, designed to perform well in the low-data regime. The proposed approach achieved a state-of-the-art performance on the BBB permeability prediction task using the BBBP dataset, surpassing existing models. With a ROC-AUC of 78.8%, the approach sets a state-of-the-art by 5.5%. We demonstrate that standard Self Attention coupled with GPS transformer performs better than other variants of attention coupled with GPS Transformer.