Despite the revolutionary breakthroughs of large-scale text-to-image diffusion models for complex vision and downstream tasks, their extremely high computational and storage costs limit their usability. Quantization of diffusion models has been explored in recent works to reduce compute costs and memory bandwidth usage. To further improve inference time, fast convolution algorithms such as Winograd can be used for convolution layers, which account for a significant portion of computations in diffusion models. However, the significant quality loss of fully quantized Winograd using existing coarser-grained post-training quantization methods, combined with the complexity and cost of finetuning the Winograd transformation matrices for such large models to recover quality, makes them unsuitable for large-scale foundation models. Motivated by the presence of a large range of values in them, we investigate the impact of finer-grained group-wise quantization in quantizing diffusion models. While group-wise quantization can largely handle the fully quantized Winograd convolution, it struggles to deal with the large distribution imbalance in a sizable portion of the Winograd domain computation. To reduce range differences in the Winograd domain, we propose finetuning only the scale parameters of the Winograd transform matrices without using any domain-specific training data. Because our method does not depend on any training data, the generalization performance of quantized diffusion models is safely guaranteed. For text-to-image generation task, the 8-bit fully-quantized diffusion model with Wino-grad provides near-lossless quality (FID and CLIP scores) in comparison to the full-precision model. This, coupled with the development of highly optimized kernels for group-wise fully quantized Winograd, improves CPU wall-clock time by 31.3% when compared to the convolution layers of a diffusion model. For image classification, our method outperforms the state-of-the-art Winograd PTQ method by 1.62% and 2.56% in top-1 ImageNet accuracy on ResNet-18 and ResNet-34, respectively, with Winograd F(6, 3).

Deep Neural Networks (DNNs) excel in learning hierarchical representations from raw data, such as images, audio, and text. To compute these DNN models with high performance and energy efficiency, these models are usually deployed onto customized hardware accelerators. Among various accelerator designs, dataflow architecture has shown promising performance due to its layer-pipelined structure and its scalability in data parallelism.Exploiting weights and activations sparsity can further enhance memory storage and computation efficiency. However, existing approaches focus on exploiting sparsity in non-dataflow accelerators, which cannot be applied onto dataflow accelerators because of the large hardware design space introduced. As such, this could miss opportunities to find an optimal combination of sparsity features and hardware designs.In this paper, we propose a novel approach to exploit unstructured weights and activations sparsity for dataflow accelerators, using software and hardware co-optimization. We propose a Hardware-Aware Sparsity Search (HASS) to systematically determine an efficient sparsity solution for dataflow accelerators. Over a set of models, we achieve an efficiency improvement ranging from $1.3 \times$ to $4.2 \times$ compared to existing sparse designs, which are either non-dataflow or non-hardware-aware. Particularly, the throughput of MobileNetV3 can be optimized to 4895 images per second. HASS is open-source: https://github.com/Yu-Zhewen/HASS

Compression techniques in machine learning (ML) independently improve a model’s inference efficiency by reducing its memory footprint while aiming to maintain its quality. This paper lays groundwork in questioning the merit of a compression pipeline involving all techniques as opposed to skipping a few by considering a case study on a keyword spotting model: DS-CNN-S. In addition, it documents improvements to the model’s training and dataset infrastructure. For this model, preliminary findings suggest that a full-scale pipeline isn’t required to achieve a competent memory footprint and accuracy, but a more comprehensive study is required.

Ever since its beginning, systems engineering has been influencing every application in our daily life. Consistent with the evolution of IT and electronics, systems design has been shaped by several structured design methodologies including Six Sigma, lean and including data analytics. Though the development of applications using systems concepts has been rapid, the focus has been limited to the fundamental methodologies leading to better system design. In this paper, a methodology, EMBED-X, is proposed for design of human-robotic systems using the concepts of systems engineering where the phases of system development are effected using Lean Six Sigma and analytics. The EMBED-X methodology encapsulates an Internet of Things (IoT) based environment. The case studies of a weather display system and development of a rover-drone are considered here. The EMBED-X framework allows flexibility in implementation and combines the principles of Six Sigma and Lean in the context of human-centered development. The human interface is accomplished using the Habermas's theory of communicative action. This methodology presents several advantages over traditional systems design. The key advantages include better coordination between human designers, design reuse, prediction of faults or failure modes and optimized cost. Therefore, the design of robotic systems, in particular, would benefit from a combined methodology, not only as a means for system development but also during training and workshops. The paper draws conclusions on the limitations of the present work and discusses the wide scope for future research.