Diffusion Transformers (DiTs) have demonstrated unprecedented performance across various generative tasks including image and video generation. However, a large amount of computations on the inference process and iterative sampling steps in the DiT models result in high computational costs, leading to substantial latency and energy consumption challenges. To address these issues, we propose a redundancy-aware DiT (RADiT), a novel software-hardware co-optimization accelerator for DiTs that minimizes redundant operations in the iterative sampling stages. We identify data redundancy by evaluating blockwise input features and skip redundant computations by reusing results from consecutive timesteps. Furthermore, to minimize accuracy degradation and maximize computational efficiency, the Dynamic Threshold Scaling Module (DTSM) and Compress and Compare Unit (CCU) are employed in the redundancy detection process. This approach enables DiTs to achieve up to $1.8 \times$ and $1.7 \times$ faster speeds for image and video generation, respectively, without compromising quality, along with 41% and 45.5% reductions in energy consumption. Our RADiT scheme improves throughput by $1.67 \times$ and $1.76 \times$ for image and video generation tasks, respectively, while maintaining output quality and significantly reducing energy consumption.

Processing-in-memory (PIM) techniques have attracted significant attention in computer architecture research due to their ability to mitigate the memory wall bottleneck for deep neural network (DNN) applications. In-memory computing (IMC) using ReRAM crossbar arrays is promising for energy efficiency but suffers from high-cost peripheral circuits, particularly analog-to-digital converters (ADCs). There are existing methods to reduce ADC energy consumption, such as lowering resolution or sharing an ADC among multiple columns. Unfortunately, these techniques reduce throughput. In this work, we propose OptiRange for PIM accelerators that enhances both energy efficiency and throughput while maintaining accuracy and model flexibility. OptiRange proposes a split-the-burden (STB) algorithm that manipulates crossbar cell values without changing the weights. STB reduces large cell values, shifts the column-sum distribution center closer to zero, and minimizes the number of low-resistance state (LRS) cells, significantly lowering the hardware-level fixed ADC resolution and improving accuracy. Furthermore, OptiRange utilizes dynamic ADC range (DAR) adaptation, determining the optimal ADC operating range per column at the software level and enabling the skipping of redundant ADC conversion cycles at runtime. An associated ADC range-based grouping (AG) strategy leverages the resulting dynamic latencies to manage synchronization and further boost system throughput. By combining weight-preserving cell manipulation with adaptive ADC operation and synchronization, OptiRange dynamically optimizes the ADC workload. Experimental results indicate that our method significantly enhances overall performance compared to conventional ReRAM-based accelerators.

Variant Angina (VA) is a condition characterized by chest pain caused by atherosclerotic spasms and contractions. It can affect younger individuals without typical risk factors and is often accompanied by symptoms like sweating, tachycardia, and low blood pressure. In severe cases, VA can result in life-threatening complications such as myocardial infarction or arrhythmia, highlighting the importance of accurate diagnosis and early prevention. Although various diagnostic methods exist in clinical practice, diagnosing VA remains difficult due to the reliance on invasive procedures that carry risks such as arterial damage and bleeding. With advancements in big data and deep learning technologies, there is growing potential for automated and accurate diagnoses through electrocardiography (ECG). This study introduces a deep learning algorithm designed specifically for diagnosing VA using ECG data. As a non-invasive approach, it offers a cost-effective and accessible solution for patients with suspected VA. To the best of our knowledge, this is the first study to apply a deep learning model solely to ECG data for VA diagnosis.

Detecting left ventricular systolic dysfunction (LVSD) traditionally relies on expensive and specialized echocardiography, limiting accessibility for many patients. To address this, researchers explore the potential of electrocardiography (ECG), a more affordable and widely available alternative, despite its historically limited performance in cardiac dysfunction detection. In this study, we present a novel approach that integrates a 1D convolutional neural networks (CNNs) with a large-scale language model (LLM) to simultaneously analyze sequential ECG data and non-sequential clinical metadata. To validate our model’s effectiveness, we conducted rigorous comparative experiments on both specially collected clinical data and public datasets, achieving an impressive AUROC of 0.97 across both. Our findings underscore the capability of ECG-based AI to accurately predict LVSD including pacemaker patients, offering a rapid and cost-effective alternative to traditional echocardiography. This innovative approach could significantly enhance early diagnosis and management of cardiac dysfunction.

BACKGROUND: Recent developments in single-cell RNA sequencing have opened up a multitude of possibilities to study tissues at the level of cellular populations. However, the heterogeneity in single-cell sequencing data necessitates appropriate procedures to adjust for technological limitations and various sources of noise when integrating datasets from different studies. While many analysis procedures employ various preprocessing steps, they often overlook the importance of selecting and optimizing the employed data transformation methods. RESULTS: This work investigates data transformation approaches used in single-cell clustering analysis tools and their effects on batch integration analysis. In particular, we compare 16 transformations and their impact on the low-dimensional representations, aiming to reduce the batch effect and integrate multiple single-cell sequencing data. Our results show that data transformations strongly influence the results of single-cell clustering on low-dimensional data space, such as those generated by UMAP or PCA. Moreover, these changes in low-dimensional space significantly affect trajectory analysis using multiple datasets, as well. However, the performance of the data transformations greatly varies across datasets, and the optimal method was different for each dataset. Additionally, we explored how data transformation impacts the analysis of deep feature encodings using deep neural network-based models, including autoencoder-based models and proto-typical networks. Data transformation also strongly affects the outcome of deep neural network models. CONCLUSIONS: Our findings suggest that the batch effect and noise in integrative analysis are highly influenced by data transformation. Low-dimensional features can integrate different batches well when proper data transformation is applied. Furthermore, we found that the batch mixing score on low-dimensional space can guide the selection of the optimal data transformation. In conclusion, data preprocessing is one of the most crucial analysis steps and needs to be cautiously considered in the integrative analysis of multiple scRNA-seq datasets.

Traffic prediction is one of the key elements to ensure the safety and convenience of citizens. Existing traffic prediction models primarily focus on deep learning architectures to capture spatial and temporal correlation. They often overlook the underlying nature of traffic. Specifically, the sensor networks in most traffic datasets do not accurately represent the actual road network exploited by vehicles, failing to provide insights into the traffic patterns in urban activities. To overcome these limitations, we propose an improved traffic prediction method based on graph convolution deep learning algorithms. We leverage human activity frequency data from National Household Travel Survey to enhance the inference capability of a causal relationship between activity and traffic patterns. Despite making minimal modifications to the conventional graph convolutional recurrent networks and graph convolutional transformer architectures, our approach achieves state-of-the-art performance without introducing excessive computational overhead.