The emergence of machine learning as a society-changing technology in the past decade has triggered concerns about people's inability to understand reasoning increasingly complex models. field IML (interpretable learning) grew out these concerns, with goal empowering various stakeholders tackle use cases, such building trust models, performing model debugging, and generally informing real human decision-making.

research-article Open Access Share on Interpretable machine learning: moving from mythos to diagnostics Authors: Valerie Chen Carnegie Mellon University, Pittsburgh, PA PAView Profile , Jeffrey Li University of Washington, Seattle, WA WAView Joon Sik Kim UniversityView Gregory Plumb Ameet Talwalkar Authors Info & Claims Communications the ACMVolume 65Issue 8August 2022 pp 43–50https://doi.org/10.1145/3546036Published:21 July 2022Publication History 0citation8,320DownloadsMetricsTotal...

AI explanations are often mentioned as a way to improve human-AI decision-making, but empirical studies have not found consistent evidence of explanations' effectiveness and, on the contrary, suggest that they can increase overreliance when system is wrong. While many factors may affect reliance support, one important factor how decision-makers reconcile their own intuition---beliefs or heuristics, based prior knowledge, experience, pattern recognition, used make judgments---with information...

Abstract One widely cited barrier to the adoption of LLMs as proxies for humans in subjective tasks is their sensitivity prompt wording—but interestingly, also display sensitivities instruction changes form response biases. We investigate extent which reflect human biases, if at all. look survey design, where biases caused by wordings “prompts” have been extensively explored social psychology literature. Drawing from these works, we design a dataset and framework evaluate whether exhibit...

Decentralized Nature-based Solutions such as Urban Green Infrastructures (UGI) are increasingly promoted to reduce flooding in urban areas. Many studies have shown the effectiveness of flood control UGI at a plot or neighbourhood level. Modelling approaches that extrapolate their reducing impact larger catchment scales often based on simplistic assumption different percentages implementation. Additionally, typically do not consider suitable space for and potential implementation constraints....

Secure Multi-Party Computation (MPC) is an area of cryptography that enables computation on sensitive data from multiple sources while maintaining privacy guarantees. However, theoretical MPC protocols often do not scale efficiently to real-world data. This project investigates the efficiency SPDZ framework, which provides implementation protocol with malicious security, in context popular machine learning (ML) algorithms. In particular, we chose applications such as linear regression and...

Social animals, including both humans and mice, are highly motivated to engage in social interactions. Short-term isolation promotes behavior, but the neural circuits through which it does so remain incompletely understood. Here, we sought identify neurons that promote behavior single-housed female exhibit increased rates of investigation, ultrasonic vocalizations (USVs), mounting during same-sex interactions follow a period short-term (3-day) isolation. We first used immunostaining for...

Evaluating in-the-wild coding capabilities of large language models (LLMs) is a challenging endeavor with no clear solution. We introduce Copilot Arena, platform to collect user preferences for code generation through native integration into developer's working environment. Arena comprises novel interface comparing pairs model outputs, sampling strategy optimized reduce latency, and prompting scheme enable completion functionality. has served over 4.5 million suggestions from 10 collected...

Maternal antibodies (MatAbs) are transferred transplacentally during pregnancy and through breast milk after birth to provide protection whilst the neonatal immune response is immature. However, MatAbs also suppress development of B cell responses via mechanisms that not well defined. can therefore result in poor vaccine performance infant, placing them at risk against potentially life-threatening pathogens such as rotavirus. It essential we understand by which interact with immunity, so...

Social animals, including both humans and mice, are highly motivated to engage in social interactions. Short-term isolation promotes behavior, but the neural circuits through which it does so remain incompletely understood. Here, we sought identify neurons that promote behavior single-housed female exhibit increased rates of investigation, ultrasonic vocalizations (USVs), mounting during same-sex interactions follow a period short-term (3 days) isolation. We first used immunostaining for...

Abstract Advances in machine learning (ML) have enabled the development of next-generation prediction models for complex computational biology problems. These developments spurred use interpretable (IML) to unveil fundamental biological insights through data-driven knowledge discovery. However, general, standards and guidelines IML usage not been well-characterized, representing a major gap toward fully realizing potential IML. Here, we introduce workflow on best practices using methods...

Summary How can managers deliver bad news with greater interactional justice? We propose a novel cognitive pathway: Construing the activity at higher (vs. lower) level increases actors' other‐oriented perspective taking, which in turn promotes enactment of justice. Three studies provide support. Studies 1 and 2 demonstrated beneficial effect construal on justice when explaining hypothetical decision. Study also showed that taking is mechanism through enactment. 3 replicated extended these...

Human β-defensin 3 (hBD3) is a cationic host defence peptide and part of the innate immune response. HBD3 present on highly copy number variable block six genes, increased associated with autoimmune disease psoriasis. It not known how this increase influences development, but psoriasis T cell-mediated activation system required for initial trigger that leads to amplification stage. We investigated effect hBD3 response primary macrophages various TLR agonists. exacerbated production type I...

Most existing evaluations of explainable machine learning (ML) methods rely on simplifying assumptions or proxies that do not reflect real-world use cases; the handful more robust settings have shortcomings in their design, generally leading to overestimation methods' utility. In this work, we seek address by conducting a study evaluates post-hoc ML setting consistent with application context and provide template for future evaluation studies. We modify improve prior e-commerce fraud...

Abstract Social animals, including both humans and mice, are highly motivated to engage in social interactions. Short-term isolation promotes behavior, but the neural circuits through which it does so remain incompletely understood. Here, we sought identify neurons that promote behavior single-housed female exhibit increased rates of investigation, ultrasonic vocalizations (USVs), mounting during same-sex interactions follow a period short-term (3-day) isolation. We first used immunostaining...

Social animals, including both humans and mice, are highly motivated to engage in social interactions. Short-term isolation increases motivation promotes behavior, but the neural circuits through which it does so remain incompletely understood. Here, we sought identify neurons that promote behavior single-housed female exhibit increased rates of investigation, ultrasonic vocalizations (USVs), mounting during same-sex interactions follow a period short-term (3-day) isolation. We first used...

A growing body of research runs human subject evaluations to study whether providing users with explanations machine learning models can help them practical real-world use cases. However, running user studies is challenging and costly, consequently each typically only evaluates a limited number different settings, e.g., often evaluate few arbitrarily selected explanation methods. To address these challenges aid design, we introduce Use-Case-Grounded Simulated Evaluations (SimEvals). SimEvals...

As machine learning (ML) pipelines affect an increasing array of stakeholders, there is a growing need for documenting how input from stakeholders recorded and incorporated. We propose FeedbackLogs, addenda to existing documentation ML pipelines, track the multiple stakeholders. Each log records important details about feedback collection process, itself, used update pipeline. In this paper, we introduce formalise process collecting FeedbackLog. also provide concrete use cases where...

Despite increasing interest in the field of Interpretable Machine Learning (IML), a significant gap persists between technical objectives targeted by researchers' methods and high-level goals consumers' use cases. In this work, we synthesize foundational work on IML evaluation into an actionable taxonomy. This taxonomy serves as tool to conceptualize researchers consumers, illustrated lack connections its cases components. It also provides foundation from which describe three-step workflow...

Machine-learning driven safety-critical autonomous systems, such as self-driving cars, must be able to detect situations where its trained model is not make a trustworthy prediction. Often viewed black-box, it non-obvious determine when will safe decision and an erroneous, perhaps life-threatening one. Prior work on novelty detection deal with highly structured data do translate well dynamic, real-world situations. This paper proposes multi-step framework for the of novel scenarios in...

Complex, multi-task problems have proven to be difficult solve efficiently in a sparse-reward reinforcement learning setting. In order sample efficient, requires reuse and sharing of low-level policies. To facilitate the automatic decomposition hierarchical tasks, we propose use step-by-step human demonstrations form natural language instructions action trajectories. We introduce dataset such crafting-based grid world. Our model consists high-level generator policy, conditioned on language....