Alliance for Safe Kids

Modern AI accelerators adopt dataflow architectures to achieve both high peak throughput (TOPS) and energy efficiency (TOPS/W). These designs feature wide datapaths and hierarchical scratchpad memories that supply dense compute arrays with high-bandwidth data access and extensive operand reuse. Complementing the compute–memory subsystem is a lightweight control path that orchestrates data movement, program loading, and register initialization. To reduce energy and area overheads, conventional processor features—such as instruction caches, execution stacks, and branch speculation—are deliberately omitted. While this streamlined design maximizes efficiency, it shifts a critical responsibility onto the compiler: transforming complex kernels into highly compact instruction streams that must fit entirely within the limited instruction buffers (IBUFFs) of the accelerator’s programmable units.In this paper, we introduce two novel compiler transformations—Loop Absorption (LA) and Loop Index Set Merging (LISM) for ultra compact code generation. Loop Absorption merges isomorphic sibling operations into a single loop body, while LISM unifies adjacent loops with similar bodies into a unified iteration space. Together, these complementary techniques eliminate redundant code patterns and produce compact hierarchical loop nests. We implement LA and LISM in the IBM Spyre compiler and evaluate them on diverse deep learning workloads including ResNet-50, Inception-v3, SSD, and BERT-Large. Across these models, our combined approach achieves a geometric mean compression of 1.48× over the baseline, enabling layers that previously exceeded IBUFF capacity to compile successfully.

In the present work, we use a merger of computational and biochemical techniques as a rational guideline for structural modification of benzofuran derivatives to find pertinent structural features for the butyrylcholinesterase inhibitory activity and selectivity. Previously, we revealed a series of 2-phenylbenzofuran compounds that displayed a selective inhibitory activity for BChE. Here, in an effort to discover novel selective BChE inhibitors with favorable physicochemical and pharmacokinetic profiles, 2-benzylbenzofurans were designed, synthesized, and evaluated as BChE inhibitors. The 2-phenylbenzofuran scaffold structure is modified by introducing one methylene spacer between the benzofuran core and the 2-phenyl ring with a hydroxyl substituent in the para or meta position. Either position 5 or 7 of the benzofuran scaffold was substituted with a bromine or chlorine atom. Further assessment of the selected list of compounds indicated that the substituent’s nature and position determined their activity and selectivity. 5-bromo-2-(4-hydroxybenzyl)benzofuran 9B proved to be the most potent butyrylcholinesterase inhibitor (IC50 = 2.93 µM) of the studied series. Computational studies were carried out to correlate the theoretical and experimental binding affinity of the compounds to the BChE protein.

End-to-end spoken language understanding (SLU) systems that process human-human or human-computer interactions are often context independent and process each turn of a conversation independently. Spoken conversations on the other hand, are very much context dependent, and dialog history contains useful information that can improve the processing of each conversational turn. In this paper, we investigate the importance of dialog history and how it can be effectively integrated into end-to-end SLU systems. While processing a spoken utterance, our proposed RNN transducer (RNN-T) based SLU model has access to its dialog history in the form of decoded transcripts and SLU labels of previous turns. We encode the dialog history as BERT embeddings, and use them as an additional input to the SLU model along with the speech features for the current utterance. We evaluate our approach on a recently released spoken dialog data set, the HarperValleyBank corpus. We observe significant improvements: 8% for dialog action and 30% for caller intent recognition tasks, in comparison to a competitive context independent end-to-end baseline system.

Walking is a complex motor function requiring coordination of all body parts. Parkinson’s disease (PD) motor signs such as rigidity, bradykinesia, and impaired balance affect movements including walking. Here, we propose a computational method to objectively assess the effects of Parkinson’s disease pathology on coordination between trunk, shoulder and limbs during the gait cycle to assess medication state and disease severity. Movements during a scripted walking task were extracted from wearable devices placed at six different body locations in participants with PD and healthy participants. Three-axis accelerometer data from each device was synchronized at the beginning of either left or right steps. Canonical templates of movements were then extracted from each body location. Movements projected on those templates created a reduced dimensionality space, where complex movements are represented as discrete values. These projections enabled us to relate the body coordination in people with PD to disease severity. Our results show that the velocity profile of the right wrist and right foot during right steps correlated with the participant’s total score on the gold standard Unified Parkinson’s Disease Rating Scale (UPRDS) with an r2 up to 0.46. Left-right symmetry of feet, trunk and wrists also correlated with the total UPDRS score with an r2 up to 0.3. In addition, we demonstrate that binary dopamine replacement therapy medication states (self-reported ‘ON’ or ‘OFF’) can be discriminated in PD participants. In conclusion, we showed that during walking, the movement of body parts individually and in coordination with one another changes in predictable ways that vary with disease severity and medication state.

In this work we propose for the first time a transformer-based framework for unsupervised representation learning of multivariate time series. Pre-trained models can be potentially used for downstream tasks such as regression and classification, forecasting and missing value imputation. By evaluating our models on several benchmark datasets for multivariate time series regression and classification, we show that not only does our modeling approach represent the most successful method employing unsupervised learning of multivariate time series presented to date, but also that it exceeds the current state-of-the-art performance of supervised methods; it does so even when the number of training samples is very limited, while offering computational efficiency. Finally, we demonstrate that unsupervised pre-training of our transformer models offers a substantial performance benefit over fully supervised learning, even without leveraging additional unlabeled data, i.e., by reusing the same data samples through the unsupervised objective.

Number of confirmed cases of COVID-19 caused by SARS-CoV-2 exceeded 5 million as of May 21, 2020. Global average of the case fatality rate of COVID-19 is about 7% so far. There exist variations in case fatality rates among countries. Particularly, Singapore and Qatar have exceptionally low case fatality rates with 0.1% while France’s rate is almost 20%. Since no magic bullet treatment for COVID-19 exists, we investigated SARS-CoV-2 strains specific to Singapore in this study to identify a clade with low pathogenicity. Variant analysis revealed that a clade with variants ORF1ab L3606F, A4489V, S2015R, T2016K, and N P13L is common in Singapore. Based on our analysis of variants and historical case statistics, the clade is dominant in a recent surge. Therefore, we suggest that low case fatality rate of Singapore possibly is attributed to the clade. Although contribution of each variant to the low pathogenicity is not clear, L3606F alone does not accomplish such low pathogenicity from the comparison with case fatality data from Japan, where L3606F is dominant. Further investigation is necessary to conclude to validate this finding.

Background Mental wellness has traditionally not been tracked by mental health clinicians. The aim is to examine the reliability and validity of the HERO Wellness Scale. Methods Eighty-four college students (age ≥18) with and without psychiatric morbidity from a small, private college in the north-central United States enrolled in the WILD 5 Wellness Program—a longitudinal, quasi-experimental clinical trial that included 2 groups (intervention and waitlist). Mental wellness scores at the beginning of the WILD 5 intervention were assessed by the HERO Wellness Scale and the World Health Organization–Five (WHO-5) Well-Being Index. Results Internal consistency was calculated from baseline data (N = 84) using Cronbach's alpha for the 5-item HERO composite = .93. The corrected item-total correlations were adequate (>.50), ranging from .67 (resilience) to .86 (mental wellness). The HERO Wellness Scale covers a representative sample of the domain of wellness. It diverges from WHO-5 items by specifically assessing for well-established positive psychology traits such as happiness, enthusiasm, resilience, and optimism. Conclusions The HERO Wellness Scale showed good validity and reliability, and should be considered for individual-level assessment of mental wellness. The HERO Wellness Scale adds to the body of knowledge in psychiatry, nursing, and psychology.

Gleason grade is a critical indicator for determining patient treatment for prostate cancer. In this paper, we analyze the viability of RNA sequencing gene expression data for Gleason grade identification. We combine datasets from the TCGA (sampled from cancer patients) and GTEx (sampled from healthy patients) databases. Using mutual information techniques, we reduce the dimensionality from 19046 genes to only the 20 most predictive genes. Then, we apply an unsupervised approach to analyze the separability of the grades of cancer. We use the t-SNE algorithm to map features into two dimensions and apply a Gaussian Mixture Model (GMM) for clustering. The result shows a clear visual separability between cancer and healthy samples. However, the grades of cancer themselves are not visually separable. Also, we apply the Mann-Whitney U test to compare the statistical similarity of the different Gleason grades and find that most grades are similar to each other. We further apply a random forest model to estimate the Gleason grade. The results show that the model accurately predicts whether a sample comes from healthy or cancer tissue. However, the model is weak in classifying the Gleason grade. The best performing model has a weighted macro-averaged F1 score of 0.66, improving on a baseline score of 0.22 obtained by random guessing. Our results indicate that the difference in gene expression among Gleason grades is relatively small compared to the difference between healthy and cancer samples. Thus, gene expression alone cannot be used for Gleason grade identification.