Shao Min Tan

We demonstrate robust and reliable signatures for the transition from quantum to classical behavior in the position probability distribution of a damped double-well system using the Qunatum State Diffusion approach to open quantum systems. We argue that these signatures are within experimental reach, for example in a doubly-clamped nanomechanical beam.

Transformer-based models, such as BERT and ViT, have achieved state-of-the-art results across different natural language processing (NLP) and computer vision (CV) tasks. However, these models are extremely memory intensive during their fine-tuning process, making them difficult to deploy on GPUs with limited memory resources. To address this issue, we introduce a new tool called SlimFit that reduces the memory requirements of these models by dynamically analyzing their training dynamics and freezing less-contributory layers during fine-tuning. The layers to freeze are chosen using a runtime inter-layer scheduling algorithm. SlimFit adopts quantization and pruning for particular layers to balance the load of dynamic activations and to minimize the memory footprint of static activations, where static activations refer to those that cannot be discarded regardless of freezing. This allows SlimFit to freeze up to 95% of layers and reduce the overall on-device GPU memory usage of transformer-based models such as ViT and BERT by an average of 2.2x, across different NLP and CV benchmarks/datasets such as GLUE, SQuAD 2.0, CIFAR-10, CIFAR-100 and ImageNet with an average degradation of 0.2% in accuracy. For such NLP and CV tasks, SlimFit can reduce up to 3.1x the total on-device memory usage with an accuracy degradation of only up to 0.4%. As a result, while fine-tuning of ViT on ImageNet and BERT on SQuAD 2.0 with a batch size of 128 requires 3 and 2 32GB GPUs respectively, SlimFit enables their fine-tuning on a single 32GB GPU without any significant accuracy degradation.

Emerging Zoned Namespace (ZNS) SSDs provide hosts with fine-grained, performance-predictable storage management. The ZNS SSD divides its address space into zones, and each zone must be written sequentially and cannot be overwritten. However, the write constraint results in frequent small modifications of file metadata to record the latest logical block addresses of the updated data blocks, increasing the overhead of fsync() system call. We comprehensively analyze the metadata overhead of fsync on a real ZNS SSD and find that (1) the update size of most metadata blocks (e.g., tens of bytes) is far less than the IO unit of the zone (e.g., 4 KiB), and frequent fsyncs cause severe metadata write amplification; (2) metadata block IOs significantly increase the fsync latency, decreasing the throughput of workloads.To reduce the metadata overhead of fsync, we build the byte-loggable zone based on the persistent write cache of the opened zone, a novel extension of ZNS, which can be appended at byte granularity instead of block. Based on the byte-loggable zone, we present BlzFS, a high-performance and crash-consistent file system for ZNS SSD. Firstly, BlzFS maintains the update ranges of the metadata block by the update bitmap. Secondly, BlzFS ensures the persistence and consistency of metadata by logging the partial updates of metadata blocks to the device’s byte-loggable zone. Finally, BlzFS restores metadata consistency by replaying the metadata update log and scanning the list of recent allocated metadata blocks. In the evaluation, BlzFS shows a 33.5%∼96.1% reduction in metadata write traffic and a 1.37×∼1.91× throughput improvement compared with the original ZNS storage system.

We conducted a study to explore participants' evaluations of different samples of sports slippers. To achieve this goal, we employed eye-tracking technology, a physiological measurement method, in conjunction with Kansei evaluation, to quantitatively analyze participants' eye-tracking data, Kansei evaluation data, and preferences for different samples. Through the analysis of the experimental results, we found that participants demonstrated a higher level of attention toward the side view of the sports slippers than others. Therefore, the side view can be considered the fundamental perspective for sports slipper research. This finding provides essential guidance for subsequent design and research endeavors. In the study of Kansei evaluation, we utilized factor analysis to reduce the dimensionality of a semantic differential scale consisting of 10 sets of adjective pairs into visual factor and affective factor. Subsequently, we explored the relationship between these two factors of sports slippers and participants' preferences. The results revealed a significant positive correlation between the visual factor and preferences. Furthermore, we identified a specific subgroup of target samples through cluster analysis. We extracted and applied design elements targeted at this subgroup based on these findings. These research findings hold significant reference value for the design and research of sports slippers and other types of footwear. By understanding participants' eye-tracking metrics and Kansei evaluation results, we can better comprehend consumers' preferences and focal points regarding sports slippers to assist designers and researchers in accurately grasping market demands during product development, thereby providing more attractive and consumer-oriented products.

Raw data for Tan et al., forest stream DOC