Search for: All records

Epithelial–mesenchymal transition (EMT) and its reverse mesenchymal–epithelial transition (MET) are critical during embryonic development, wound healing and cancer metastasis. While phenotypic changes during short-term EMT induction are reversible, long-term EMT induction has been often associated with irreversibility. Here, we show that phenotypic changes seen in MCF10A cells upon long-term EMT induction by TGF β need not be irreversible, but have relatively longer time scales of reversibility than those seen in short-term induction. Next, using a phenomenological mathematical model to account for the chromatin-mediated epigenetic silencing of the miR-200 family by ZEB family, we highlight how the epigenetic memory gained during long-term EMT induction can slow the recovery to the epithelial state post-TGF β withdrawal. Our results suggest that epigenetic modifiers can govern the extent and time scale of EMT reversibility and advise caution against labelling phenotypic changes seen in long-term EMT induction as ‘irreversible’. 

Approximate computing is a promising way to improve the power efficiency of deep learning. While recent work proposes new arithmetic circuits (adders and multipliers) that consume substantially less power at the cost of computation errors, these approximate circuits decrease the end-to-end accuracy of common models. We present AutoApprox, a framework to automatically generate approximate low-power deep learning accelerators without any accuracy loss. AutoApprox generates a wide range of approximate ASIC accelerators with a TPUv3 systolic-array template. AutoApprox uses a learned router to assign each DNN layer to an approximate systolic array from a bank of arrays with varying approximation levels. By tailoring this routing for a specific neural network architecture, we discover circuit designs without the accuracy penalty from prior methods. Moreover, AutoApprox optimizes for the end-to-end performance, power and area of the the whole chip and PE mapping rather than simply measuring the performance of the arithmetic units in iso-lation. To our knowledge, our work is the first to demonstrate the effectiveness of custom-tailored approximate circuits in delivering significant chip-level energy savings with zero accuracy loss on a large-scale dataset such as ImageNet. AutoApprox synthesizes a novel approximate accelerator based on the TPU that reduces end-to-end power consumption by 3.2% and area by 5.2% at a sub-10nm process with no degradation in ImageNet validation top-1 and top-5 accuracy. 

A sanitized drinking water supply is an unconditional requirement for public health and the overall prosperity of humanity. Potential microbial and chemical contaminants of drinking water have been identified by a joint effort between the World Health Organization (WHO) and the United Nations Children’s Fund (UNICEF), who together establish guidelines that define, in part, that the presence of Escherichia coli (E. coli) in drinking water is an indication of inadequate sanitation and a significant health risk. As E. coli is a nearly ubiquitous resident of mammalian gastrointestinal tracts, no detectable counts in 100 mL of drinking water is the standard used worldwide as an indicator of sanitation. The currently accepted EPA method relies on filtration, followed by growth on selective media, and requires 24–48 h from sample to results. In response, we developed a rapid bacteriophage-based detection assay with detection limit capabilities comparable to traditional methods in less than a quarter of the time. We coupled membrane filtration with selective enrichment using genetically engineered bacteriophages to identify less than 20 colony forming units (CFU) E. coli in 100 mL drinking water within 5 h. The combination of membrane filtration with phage infection produced a novel assay that demonstrated a rapid, selective, and sensitive detection of an indicator organism in large volumes of drinking water as recommended by the leading world regulatory authorities.