Human perceptual decisions are critically important. They can be socially significant, life-changing or even catastrophic if wrong, such as decisions made by radiologists, Olympic judges, TSA screeners, or vehicle operators. Many of these crucial judgments are made based on the perceptual expertise of individual observers. However, not everyone possesses the same level of expertise, even among professionals. There are two aspects that characterize the individual differences in expertise: variations in sensitivity and variations in bias. Past research has thoroughly investigated differences in sensitivity across many domains. However, the possibility of variations in perceptual bias has been largely ignored.

Here, we filled this gap by investigating the idiosyncratic variations in human visual perceptual biases. In the first study, we identified and characterized the unique perceptual biases in position perception and showed that these biases might originate from variations in visual acuity and they can influence the perceived appearance of objects, suggesting that idiosyncratic perceptual biases could propagate through different levels of visual processing and be associated with each other. To understand whether perceptual biases can have a real-world impact, in the second study, we further tested radiologists with artificial and realistic medical images and found that even expert radiologists were characterized with unique and systematic perceptual biases towards medical images. Given the prevalence of perceptual biases, the third study aimed to discover potential constructive ways to utilize the biases. We showed that when multiple observers make a collective decision, grouping individuals who have uncorrelated and thus independent perceptual biases can improve their combined performance.

Together, these results highlighted the crucial importance of measuring and understanding individual perceptual biases and they have widespread implications for many real-world professions and situations such as driving, radiological or TSA screening and Olympic judges.

Heterogeneous systems emerge to enhance application performance by distributing tasks across various domain-specific accelerators and memory devices. However, adopting such systems often results in unanticipated performance characteristics in modern workloads, unforeseen vulnerabilities introduced by new architecture designs, and programming challenges utilizing novel architectures.

This dissertation explores a cross-stack approach to building heterogeneous systems. We start from the architecture level, studying the performance characteristics of emerging heterogeneous memory architectures. This led to our system-level study integrating diverse types of memory devices at a large scale, which facilitates efficient training of large machine learning models. These explorations revealed vulnerabilities in emerging memory architectures and led to our development of architectural side-channel attacks and cross-stack defenses to secure heterogeneous systems. Finally, to facilitate the adoption of heterogeneous systems, we proposed a novel programming model that provides a unified interface for heterogeneous computing.

The low-density lipoprotein receptor-related protein-1 (LRP1) is an endocytic and cell signaling receptor, highly involved in innate immunity. Although activation of membrane-bound LRP1 results in anti-inflammatory activity in vitro and in vivo, and genetic deletion of LRP1 in Schwann cells increase neuropathic pain, the ability of LRP1 agonists to alleviate pain is unknown. Thus, we tested whether a novel small peptide, SP16, derived from alpha-1-anti-trypsin and binds LRP1 with high affinity, could regulate pain states in three distinct mice models. Initially, we determined that SP16 robustly activates cell signaling in PC12 cells via an LRP1 dependent manner. SP16 also increased neurite length and expression of regenerative genes in cultured adult primary sensory neurons. In response to intraplantar capsaicin, SP16, two SP16 analogs (A2-5, A15) and a well-known LRP1 activator, enzymatically inactive (EI)-tissue plasminogen activator (tPA) significantly reduced paw licking time compared to vehicle. SP16 blocked the development of tactile allodynia after partial sciatic nerve ligation (P<0.05). The underlying mechanism includes limited recruitment of CD11b+ inflammatory cells (P<0.01) into the nerve injury site and reduced levels of Toll-like receptor-4 (TLR4) (P<0.05). In injured DRGs, SP16 reduced the presence of CD11b+ cells and GFAP, suggesting that inflammatory and satellite cell activation was inhibited. Following intraplantar formalin injection, SP16-treated mice showed reduced responses in phase 1 (P<0.01) and delayed the onset of phase 2 (P<0.05). We conclude that SP16 inhibits acute nociceptive, inflammatory, and neuropathic pain-related behaviors, suggesting that LRP1 agonists have uniquely broad anti-inflammatory activities that minimize the development of pain states.