arXiv subjects

William Paul

Publications and source records attributed to William Paul.

30 records · Page 2Linked to original sources

arXiv2020

Addressing Artificial Intelligence Bias in Retinal Disease Diagnostics

This study evaluated generative methods to potentially mitigate AI bias when diagnosing diabetic retinopathy (DR) resulting from training data imbalance, or domain generalization which occurs when deep learning systems (DLS) face concepts at test/inference time they were not initially trained on. The public domain Kaggle-EyePACS dataset (88,692 fundi and 44,346 individuals, originally diverse for ethnicity) was modified by adding clinician-annotated labels and constructing an artificial scenario of data imbalance and domain generalization by disallowing training (but not testing) exemplars for images of retinas with DR warranting referral (DR-referable) and from darker-skin individuals, who presumably have greater concentration of melanin within uveal melanocytes, on average, contributing to retinal image pigmentation. A traditional/baseline diagnostic DLS was compared against new DLSs that would use training data augmented via generative models for debiasing. Accuracy (95% confidence intervals [CI]) of the baseline diagnostics DLS for fundus images of lighter-skin individuals was 73.0% (66.9%, 79.2%) vs. darker-skin of 60.5% (53.5%, 67.3%), demonstrating bias/disparity (delta=12.5%) (Welch t-test t=2.670, P=.008) in AI performance across protected subpopulations. Using novel generative methods for addressing missing subpopulation training data (DR-referable darker-skin) achieved instead accuracy, for lighter-skin, of 72.0% (65.8%, 78.2%), and for darker-skin, of 71.5% (65.2%,77.8%), demonstrating closer parity (delta=0.5%) in accuracy across subpopulations (Welch t-test t=0.111, P=.912). Findings illustrate how data imbalance and domain generalization can lead to disparity of accuracy across subpopulations, and show that novel generative methods of synthetic fundus images may play a role for debiasing AI.

cs.AI↗

arXiv2020

AI Progress in Skin Lesion Analysis

We examine progress in the use of AI for detecting skin lesions, with particular emphasis on the erythema migrans rash of acute Lyme disease, and other lesions, such as those from conditions like herpes zoster (shingles), tinea corporis, erythema multiforme, cellulitis, insect bites, or tick bites. We discuss important challenges for these applications, in particular the problems of AI bias regarding the lack of skin images in dark skinned individuals, being able to accurately detect, delineate, and segment lesions or regions of interest compared to normal skin in images, and low shot learning (addressing classification with a paucity of training images). Solving these problems ranges from being highly desirable requirements -- e.g. for delineation, which may be useful to disambiguate between similar types of lesions, and perform improved diagnostics -- or required, as is the case for AI de-biasing, to allow for the deployment of fair AI techniques in the clinic for skin lesion analysis. For the problem of low shot learning in particular, we report skin analysis algorithms that gracefully degrade and still perform well at low shots, when compared to baseline algorithms: when using a little as 10 training exemplars per class, the baseline DL algorithm performance significantly degrades, with accuracy of 56.41%, close to chance, whereas the best performing low shot algorithm yields an accuracy of 85.26%.

eess.IV↗

arXiv2019

Hierarchical Variational Imitation Learning of Control Programs

Autonomous agents can learn by imitating teacher demonstrations of the intended behavior. Hierarchical control policies are ubiquitously useful for such learning, having the potential to break down structured tasks into simpler sub-tasks, thereby improving data efficiency and generalization. In this paper, we propose a variational inference method for imitation learning of a control policy represented by parametrized hierarchical procedures (PHP), a program-like structure in which procedures can invoke sub-procedures to perform sub-tasks. Our method discovers the hierarchical structure in a dataset of observation-action traces of teacher demonstrations, by learning an approximate posterior distribution over the latent sequence of procedure calls and terminations. Samples from this learned distribution then guide the training of the hierarchical control policy. We identify and demonstrate a novel benefit of variational inference in the context of hierarchical imitation learning: in decomposing the policy into simpler procedures, inference can leverage acausal information that is unused by other methods. Training PHP with variational inference outperforms LSTM baselines in terms of data efficiency and generalization, requiring less than half as much data to achieve a 24% error rate in executing the bubble sort algorithm, and to achieve no error in executing Karel programs.

cs.LG↗

arXiv2019

Tuning the exchange bias on a single atom from 1 mT to 10 T

Shrinking spintronic devices to the nanoscale ultimately requires localized control of individual atomic magnetic moments. At these length scales, the exchange interaction plays important roles, such as in the stabilization of spin-quantization axes, the production of spin frustration, and creation of magnetic ordering. Here, we demonstrate the precise control of the exchange bias experienced by a single atom on a surface, covering an energy range of four orders of magnitude. The exchange interaction is continuously tunable from milli-eV to micro-eV by adjusting the separation between a spin-1/2 atom on a surface and the magnetic tip of a scanning tunneling microscope (STM). We seamlessly combine inelastic electron tunneling spectroscopy (IETS) and electron spin resonance (ESR) to map out the different energy scales. This control of exchange bias over a wide span of energies provides versatile control of spin states, with applications ranging from precise tuning of quantum state properties, to strong exchange bias for local spin doping. In addition we show that a time-varying exchange interaction generates a localized AC magnetic field that resonantly drives the surface spin. The static and dynamic control of the exchange interaction at the atomic-scale provides a new tool to tune the quantum states of coupled-spin systems.

cond-mat.mes-hall↗

arXiv2018

Ray: A Distributed Framework for Emerging AI Applications

The next generation of AI applications will continuously interact with the environment and learn from these interactions. These applications impose new and demanding systems requirements, both in terms of performance and flexibility. In this paper, we consider these requirements and present Ray---a distributed system to address them. Ray implements a unified interface that can express both task-parallel and actor-based computations, supported by a single dynamic execution engine. To meet the performance requirements, Ray employs a distributed scheduler and a distributed and fault-tolerant store to manage the system's control state. In our experiments, we demonstrate scaling beyond 1.8 million tasks per second and better performance than existing specialized systems for several challenging reinforcement learning applications.

cs.DC↗

arXiv2017

Engineering the eigenstates of coupled spin-1/2 atoms on a surface

Quantum spin networks having engineered geometries and interactions are eagerly pursued for quantum simulation and access to emergent quantum phenomena such as spin liquids. Spin-1/2 centers are particularly desirable because they readily manifest coherent quantum fluctuations. Here we introduce a controllable spin-1/2 architecture consisting of titanium atoms on a magnesium oxide surface. We tailor the spin interactions by atomic-precision positioning using a scanning tunneling microscope (STM), and subsequently perform electron spin resonance (ESR) on individual atoms to drive transitions into and out of quantum eigenstates of the coupled-spin system. Interactions between the atoms are mapped over a range of distances extending from highly anisotropic dipole coupling, to strong exchange coupling. The local magnetic field of the magnetic STM tip serves to precisely tune the superposition states of a pair of spins. The precise control of the spin-spin interactions and ability to probe the states of the coupled-spin network by addressing individual spins will enable exploration of quantum many-body systems based on networks of spin-1/2 atoms on surfaces.

cond-mat.mes-hall↗

arXiv2017

Real-Time Machine Learning: The Missing Pieces

Machine learning applications are increasingly deployed not only to serve predictions using static models, but also as tightly-integrated components of feedback loops involving dynamic, real-time decision making. These applications pose a new set of requirements, none of which are difficult to achieve in isolation, but the combination of which creates a challenge for existing distributed execution frameworks: computation with millisecond latency at high throughput, adaptive construction of arbitrary task graphs, and execution of heterogeneous kernels over diverse sets of resources. We assert that a new distributed execution framework is needed for such ML applications and propose a candidate approach with a proof-of-concept architecture that achieves a 63x performance improvement over a state-of-the-art execution framework for a representative application.

cs.DC↗

arXiv2016

Reading and Writing Single-Atom Magnets

The highest-density magnetic storage media will code data in single-atom bits. To date, the smallest individually addressable bistable magnetic bits on surfaces consist of 5-12 atoms. Long magnetic relaxation times were demonstrated in molecular magnets containing one lanthanide atom, and recently in ensembles of single holmium (Ho) atoms supported on magnesium oxide (MgO). Those experiments indicated the possibility for data storage at the fundamental limit, but it remained unclear how to access the individual magnetic centers. Here we demonstrate the reading and writing of individual Ho atoms on MgO, and show that they independently retain their magnetic information over many hours. We read the Ho states by tunnel magnetoresistance and write with current pulses using a scanning tunneling microscope. The magnetic origin of the long-lived states is confirmed by single-atom electron paramagnetic resonance (EPR) on a nearby Fe sensor atom, which shows that Ho has a large out-of-plane moment of $(10.1 \pm 0.1)$ $μ_{\rm B}$ on this surface. In order to demonstrate independent reading and writing, we built an atomic scale structure with two Ho bits to which we write the four possible states and which we read out remotely by EPR. The high magnetic stability combined with electrical reading and writing shows that single-atom magnetic memory is possible.

cond-mat.mes-hall↗

arXiv2013

Transient adhesion and conductance phenomena in initial nanoscale mechanical contacts between dissimilar metals

We report on transient adhesion and conductance phenomena associated with tip wetting in mechanical contacts produced by the indentation of a clean W(111) tip into a Au(111) surface. A combination of atomic force microscopy and scanning tunneling microscopy was used to carry out indentation and to image residual impressions in ultra-high vacuum. The ~7 nm radii tips used in these experiments were prepared and characterized by field ion microscopy in the same instrument. The very first indentations of the tungsten tips show larger conductance and pull-off adhesive forces than subsequent indentations. After ~30 indentations to a depth of ~1.7 nm, the maximum conductance and adhesion forces reach steady-state values approximately 12x and 6x smaller than their initial value. Indentation of W(111) tips into Cu(100) was also performed to investigate the universality of tip wetting phenomena with a different substrate. We propose a model from contact mechanics considerations which quantitatively reproduces the observed decay rate of the conductance and adhesion drops with a 1/e decay constant of 9-14 indentation cycles. The results show that the surface composition of an indenting tip plays an important role in defining the mechanical and electrical properties of indentation contacts.

cond-mat.mtrl-sci↗

arXiv2013

FIM tips in SPM: Apex orientation and temperature considerations on atom transfer and diffusion

Atoms transferred to W(111) and W(110) tip apices from the Au(111) surface during tunneling and approach to mechanical contact experiments in STM are characterized in FIM at room temperature and at 158 K. The different activation energies for diffusion on the (111) and (110) tip planes and the experiment temperature are shown to be important factors controlling the extent of changes to the atomic structure of the tip. W(111) tips are much better suited to scanning probe studies which require the characterization of an atomically defined tip and subsequent verification of its integrity in FIM. The statistics of the observed spikes in the tunneling current when the tips are approached to Au(111) are interpreted using a simple model of adatoms diffusing through the STM junction.

cond-mat.mtrl-sci↗

arXiv2013

Comment on 'Field ion microscopy characterized tips in noncontact atomic force microscopy: Quantification of long-range force interactions'

A recent article by Falter et al. (Phys. Rev. B 87, 115412 (2013)) presents experimental results using field ion microscopy characterized tips in noncontact atomic force microscopy in order to characterize electrostatic and van der Waals long range forces. In the article, the tip radius was substantially underestimated at ~4.7 nm rather than ~8.1 nm due to subtleties in the application of the ring counting method. We point out where common errors in ring counting arise in order to benefit future experimental work in which the determination of tip radius by FIM is important.

cond-mat.mes-hall↗

arXiv2012

Implementation of atomically defined Field Ion Microscopy tips in Scanning Probe Microscopy

The Field Ion Microscope (FIM) can be used to characterize the atomic configuration of the apex of sharp tips. These tips are well suited for Scanning Probe Microscopy (SPM) since they predetermine SPM resolution and electronic structure for spectroscopy. A protocol is proposed to preserve the atomic structure of the tip apex from etching due to gas impurities during the transfer period from FIM to SPM, and estimations are made regarding the time limitations of such an experiment due to contamination by ultra-high vacuum (UHV) rest gases. While avoiding any current setpoint overshoot to preserve the tip integrity, we present results from approaches of atomically defined tungsten tips to the tunneling regime with Au(111), HOPG, and Si(111) surfaces at room temperature. We conclude from these experiments that adatom mobility and physisorbed gas on the sample surface limit the choice of surfaces for which the tip integrity is preserved in tunneling experiments at room temperature. The atomic structure of FIM tip apices is unchanged only after tunneling to the highly reactive Si(111) surface.

cond-mat.mes-hall↗