Machine learning:Trends, perspectives, and prospects

1 Despite practical challenges, we are hopeful thatinformeddiscussionsamongpolicy-maker sandthepublic about data and the capabilities of machinelearning, will lead to insightful designs of programsand policies that can balance the goals of protectingprivacy and ensuring fairness with those of reapingthe benefits to scientific research and to individualand public health. Our commitments to privacy andfairness are evergreen, but our policy choices mustadapt to advance them, and support new tech-niques for deepening our AND NOTES1. M. De Choudhury, S. Counts, E. Horvitz, A. Hoff, inProceedingsof International Conference on Weblogs and Social Media[Association for the Advancement of Artificial Intelligence(AAAI), Palo Alto, CA, 2014].2. J. S. Brownstein, C. C. Freifeld, L. C. Madoff,N. Engl. J. , 2153 2155 (2009).3. G. Eysenbach,J. Med. Internet , e11 (2009).4. D. A. Broniatowski, M. J. Paul, M. Dredze,PLOS ONE8, e83672(2013).

2 5. A. Sadilek, H. Kautz, V. Silenzio, inProceedings of theTwenty-Sixth AAAI Conference on Artificial Intelligence(AAAI, Palo Alto, CA, 2012).6. M. De Choudhury, S. Counts, E. Horvitz, inProceedings of theSIGCHIC onference on Human Factors in Computing Systems(Association for Computing Machinery, New York, 2013),pp. 3267 R. W. White, R. Harpaz, N. H. Shah, W. DuMouchel, E. Horvitz,Clin. Pharmacol. , 239 246 (2014).8. Samaritans Radar; Shut down Samaritans Radar; Equal Employment Opportunity Commission (EEOC), 29 Code of Federal Regulations ( ), (g) (2013).11. EEOC, 29 CFR (c) (2013).12. M. A. Rothstein,J. Law Med. Ethics36, 837 840 (2008).13. Executive Office of the President,Big Data: SeizingOpportunities, Preserving Values(White House, Washington,DC, 2014); Letter from Maneesha Mithal, FTC, to Reed Freeman, Morrison,& Foerster LLP, Counsel for Netflix, 2 [closing letter] (2010); In re Facebook, Complaint, FTC File No.

3 092 3184 (2012).16. FTC Staff Report,Mobile Privacy Disclosures: Building Trust ThroughTransparency(FTC,Washington,DC,20 13); FTC,Protecting Consumer Privacy in an Era of Rapid Change:Recommendations for Businesses and Policymakers(FTC,Washington, DC, 2012).18. Directive 95/46/ec of the European Parliament and of TheCouncil of Europe, 24 October L. Sweeney, Online ads roll the dice [blog]; FTC, Big data: A tool for inclusion or exclusion? (workshop,FTC, Washington, DC, 2014); FTC,Data Brokers: A Call for Transparency and Accountability(FTC, Washington, DC, 2014); J. Podesta, Big data and privacy: 1 year out [blog]; White House Council of Economic Advisers,Big Data andDifferential Pricing(White House, Washington, DC, 2015).24. Executive Office of the President,Big Data and Differential Processing(White House, Washington, DC, 2015); Executive Office of the President,Big Data: SeizingOpportunities, Preserving Values(White House, Washington,DC, 2014); President s Council of Advisors on Science and Technology(PCAST),Big Data and Privacy: A Technological Perspective(White House, Washington, DC, 2014); European Commission, Proposal for a Regulation of the EuropeanParliament and of the Council on the Protection of Individualswith regard to the processing of personal data and on the freemovement of such data (General Data Protection Regulation),COM(2012) 11 final (2012); Ireland Limited, J.

4 Unlawful datatransmission to the ( PRISM ), 166 and 167 (2013); learning: trends ,perspectives, and prospectsM. I. Jordan1*and T. M. Mitchell2* Machine learning addresses the question of how to build computers that improveautomatically through experience. It is one of today s most rapidly growing technical fields,lying at the intersection of computer science and statistics, and at the core of artificialintelligence and data science. Recent progress in Machine learning has been driven both bythe development of new learning algorithms and theory and by the ongoing explosion in theavailability of online data and low-cost computation. The adoption of data-intensivemachine-learning methods can be found throughout science, technology and commerce,leading to more evidence-based decision-making across many walks of life, includinghealth care, manufacturing, education, financial modeling, policing, and learning is a discipline focusedon two interrelated questions: How canone construct computer systems that auto-matically improve through experience?

5 And What are the fundamental statistical-computational-information-th eoretic laws thatgovern all learning systems, including computers,humans, and organizations? The study of machinelearning is important both for addressing thesefundamental scientific and engineering ques-tions and for the highly practical computer learning has progressed dramati-cally over the past two decades, from laboratorycuriosity to a practical technology in widespreadcommercial use. Within artificial intelligence (AI), Machine learning has emerged as the methodof choice for developing practical software forcomputer vision, speech recognition, natural lan-guage processing, robot control, and other ap-plications. Many developers of AI systems nowrecognize that, for many applications, it can befar easier to train a system by showing it exam-ples of desired input-output behavior than toprogram it manually by anticipating the desiredresponse for all possible inputs.

6 The effect of ma-chine learning has also been felt broadly acrosscomputer science and across a range of indus-tries concerned with data-intensive issues, suchas consumer services, the diagnosis of faults incomplex systems, and the control of logisticschains. There has been a similarly broad range ofeffects across empirical sciences, from biology tocosmology to social science, as Machine -learningmethods have been developed to analyze high-throughput experimentaldata in novel ways. SeeFig. 1 for a depiction of some recent areas of ap-plication of Machine learning problem can be defined as theproblem of improving some measure of perform-ance when executing some task, through sometype of training experience. For example, in learn-ing to detect credit-card fraud, the task is to as-sign a label of fraud or not fraud to any givencredit-card transaction. The performance metricto be improved might be the accuracy of thisfraud classifier, and the training experience mightconsist of a collection of historical credit-cardtransactions, each labeled in retrospect as fraud-ulent or not.

7 Alternatively, one might define adifferent performance metric that assigns a higherpenalty when fraud is labeled not fraud thanwhen not fraud is incorrectly labeled fraud. One might also define a different type of trainingexperience for example, by including unlab-eled credit-card transactions along with diverse array of Machine -learning algorithmshas been developed to cover the wide variety ofdata and problem types exhibited across differ-ent Machine -learning problems (1,2). Conceptual-ly, Machine -learning algorithms can be viewed assearching through a large space of candidateprograms, guided by training experience, to finda program that optimizes the performance algorithms vary greatly, in partby the way in which they represent candidateprograms ( , decision trees, mathematical func-tions, and general programming languages) and inpartbythewayinwhichtheysearchthroughth isspace of programs ( , optimization algorithmswith well-understood convergence guaranteesand evolutionary search methods that evaluatesuccessivegenerationsofrandomlym utatedpro-grams).

8 Here, we focus on approaches that havebeen particularly successful to algorithms focus on function approxi-mation problems, where the task is embodiedin a function ( , given an input transaction, out-put a fraud or not fraud label), and the learn-ing problem is to improve the accuracy of thatfunction, with experience consisting of a sampleof known input-output pairs of the function. Insome cases, the function is represented explicit-ly as a parameterized functional form; in othercases, the function is implicit and obtained via asearch process, a factorization, an JULY 2015 VOL 349 ISSUE 62452551 Department of Electrical Engineering and ComputerSciences, Department of Statistics, University of California,Berkeley, CA, Learning Department, CarnegieMellon University, Pittsburgh, PA, USA.*Corresponding author. E-mail: ( ) on July 20, from on July 20, from on July 20, from on July 20, from on July 20, from on July 20, from procedure, or a simulation-based procedure.

9 Evenwhen implicit, the function generally dependson parameters or other tunable degrees of free-dom,andtrainingcorrespondstofinding valuesfor these parameters that optimize the perform-ance the learning algorithm, a key scien-tific and practical goal is to theoretically character-ize the capabilities of specific learning algorithmsand the inherent difficulty of any given learningproblem: How accurately can the algorithm learnfrom a particular type and volume of trainingdata? How robust is the algorithm to errors in itsmodeling assumptions or to errors in the train-ing data? Given a learning problem with a givenvolume of training data, is it possible to design asuccessful algorithm or is this learning problemfundamentally intractable? Such theoretical char-acterizations of Machine -learning algorithms andproblems typically make use of the familiar frame-works of statistical decision theory and compu-tational complexity theory.

10 In fact, attempts tocharacterize Machine -learning algorithms the-oretically have led to blends of statistical andcomputationaltheoryinwhichthegoalisto simul-taneously characterize the sample complexity(how much data are required to learn accurately)and the computational complexity (how muchcomputation is required) and to specify how thesedepend on features of the learning algorithm suchas the representation it uses for what it learns(3 6). A specific form of computational analysisthat has proved particularly useful in recentyears has been that of optimization theory,with upper and lower bounds on rates of con-vergence of optimization procedures mergingwell with the formulation of Machine -learningproblems as the optimization of a performancemetric (7,8).As a field of study, Machine learning sits at thecrossroads of computer science, statistics and avariety of other disciplines concerned with auto-matic improvement overtime, and inference anddecision-making under uncertainty.