Meta tags:
description= Cofounder and CEO of Elicit. Building AI to help people reason well about hard problems.;
Headings (most frequently used words):
andreas, stuhlmüller, interests, writing, talks, publications, projects, elsewhere, epistemics, transformative, ai, foundations,
Text of the page (most frequently used words):
and (293), the (235), that (92), for (88), probabilistic (78), #andreas (71), this (70), learning (61), models (51), programs (49), are (49), inference (46), with (42), stuhlmüller (40), goodman (40), human (40), noah (39), how (37), can (37), reasoning (36), programming (36), org (35), author (34), title (34), year (34), https (34), bibtex (34), model (34), abstract (34), language (33), url (32), from (32), stuhlm (30), based (29), ller (29), our (29), about (29), systems (28), cognitive (27), these (24), papers (22), people (22), science (21), stuhlmueller (20), program (20), conference (20), their (20), research (19), keywords (19), more (19), agents (19), booktitle (18), proceedings (18), process (18), tasks (18), inproceedings (17), bayesian (17), arxiv (17), questions (17), they (17), generative (16), languages (16), pdf (16), algorithms (16), such (16), which (16), using (15), cognition (15), world (14), artificial (14), intelligence (14), describe (14), what (14), show (14), information (14), owain (14), evans (14), judgments (14), elicit (13), make (13), algorithm (13), use (13), 2012 (13), have (13), theory (13), not (13), task (13), ast (12), structured (12), approach (12), annual (12), concepts (12), examples (12), modeling (12), find (12), physical (12), work (12), 2016 (12), reinforcement (12), 2018 (12), mcmc (11), stochastic (11), tenenbaum (11), society (11), into (11), distribution (11), arguments (11), will (11), some (11), agent (10), one (10), new (10), given (10), ndg (10), two (10), nips (10), blame (10), when (10), implicature (10), help (10), good (10), dynamic (10), factored (10), inverses (10), 2015 (10), representations (9), problem (9), answer (9), question (9), thirty (9), second (9), approaches (9), lightweight (9), used (9), abs (9), report (9), understanding (9), workshop (9), training (9), but (9), preferences (9), future (9), ought (9), evidence (8), learn (8), like (8), joshua (8), complex (8), 2011 (8), choices (8), fast (8), machine (8), only (8), technical (8), each (8), knowledge (8), between (8), conditioning (8), other (8), physics (8), planning (8), dialog (8), dataset (8), deliberative (8), byun (8), jungwon (8), 2023 (8), source (7), decisions (7), compositionality (7), simple (7), real (7), than (7), objects (7), efficient (7), data (7), then (7), ullman (7), outcome (7), social (7), distributions (7), steps (7), 2013 (7), has (7), humans (7), misc (7), where (7), coarse (7), fine (7), without (7), evaluating (7), implementation (6), building (6), compositional (6), thesis (6), sampling (6), reason (6), over (6), makes (6), demonstrate (6), observations (6), many (6), embeddings (6), recent (6), structure (6), predictions (6), david (6), international (6), method (6), existing (6), illustrate (6), nonstandard (6), interpretations (6), article (6), within (6), functions (6), nested (6), john (6), tomer (6), why (6), framework (6), results (6), intervention (6), value (6), state (6), test (6), solve (6), inverse (6), scenes (6), intuitive (6), most (6), also (6), might (6), scheme (6), level (6), ask (6), loop (6), markets (6), even (6), set (6), few (6), predicting (6), academic (6), raft (6), 2021 (6), 2022 (6), decomposition (6), publications (5), church (5), high (5), space (5), order (5), recursive (5), capture (5), processes (5), experiment (5), compare (5), well (5), execution (5), enables (5), neural (5), techniques (5), special (5), archiveprefix (5), eprint (5), primaryclass (5), all (5), introduce (5), number (5), domain (5), view (5), suggest (5), case (5), formalize (5), invert (5), apply (5), act (5), marginal (5), caching (5), game (5), however (5), been (5), 2014 (5), environments (5), multiple (5), time (5), short (5), yet (5), systematic (5), book (5), bounded (5), ability (5), end (5), catastrophes (5), sentence (5), performance (5), datasets (5), classification (5), benchmark (5), workflow (5), 2026 (5), open (4), researchers (4), institute (4), importance (4), both (4), adaptive (4), monte (4), carlo (4), representation (4), variety (4), wingate (4), statistics (4), via (4), implement (4), advances (4), processing (4), because (4), design (4), represented (4), computation (4), different (4), them (4), proposals (4), merging (4), patterns (4), previous (4), search (4), consider (4), reducing (4), including (4), sub (4), cogsci2012 (4), fourth (4), attribution (4), much (4), first (4), represent (4), prior (4), further (4), scalar (4), evaluate (4), listeners (4), speaker (4), implies (4), rational (4), computational (4), get (4), propose (4), metropolis (4), hastings (4), journal (4), range (4), beliefs (4), explore (4), challenges (4), applied (4), support (4), dynamical (4), explain (4), properties (4), howpublished (4), natural (4), explains (4), required (4), provide (4), due (4), three (4), hawkins (4), robert (4), you (4), answers (4), answerer (4), 1509 (4), daniel (4), slow (4), aaai (4), preference (4), biases (4), optimal (4), salvatier (4), interactive (4), improve (4), 2017 (4), william (4), saunders (4), trial (4), supervised (4), baseline (4), parameters (4), understand (4), built (4), nli (4), valid (4), better (4), zachary (4), rachbach (4), ben (4), 2020 (4), text (4), shot (4), reppert (4), justin (4), stebbing (4), luke (4), outcomes (4), think (4), charlie (4), george (4), iterated (4), hallucination (4), hallucinations (4), enough (4), updates (3), com (3), blog (3), google (3), projects (3), mit (3), implementations (3), web (3), tools (3), synthesize (3), stakes (3), university (3), induction (3), could (3), psychology (3), goal (3), tree (3), class (3), under (3), markov (3), chain (3), methods (3), defined (3), terms (3), concept (3), constructing (3), generated (3), judgements (3), transforming (3), random (3), values (3), additional (3), code (3), development (3), functional (3), interpretation (3), dependencies (3), while (3), 1110 (3), 5667 (3), abstraction (3), posterior (3), counterfactual (3), variable (3), assign (3), change (3), would (3), choose (3), asked (3), assume (3), approximately (3), interaction (3), experiments (3), speech (3), direct (3), system (3), equations (3), teaching (3), samples (3), intermediate (3), sampler (3), doi (3), inferences (3), way (3), uncertainty (3), networks (3), amortized (3), setting (3), factorization (3), before (3), uses (3), basic (3), early (3), throughout (3), life (3), hierarchical (3), levels (3), expressive (3), challenging (3), able (3), perform (3), line (3), top (3), down (3), feature (3), ppls (3), interest (3), continuations (3), technology (3), just (3), develop (3), showing (3), providing (3), large (3), informative (3), increasingly (3), questioner (3), constant (3), distinguish (3), interests (3), 02962 (3), require (3), its (3), optimality (3), incrementalized (3), present (3), inconsistent (3), should (3), shown (3), person (3), utility (3), decision (3), making (3), often (3), false (3), may (3), deviations (3), agnostic (3), machines (3), expert (3), advice (3), reward (3), self (3), answering (3), requires (3), asker (3), did (3), possible (3), aligned (3), towards (3), safe (3), learned (3), avoid (3), serious (3), hours (3), instead (3), trained (3), infersent (3), improving (3), automating (3), difficult (3), involves (3), current (3), step (3), invalid (3), participant (3), argument (3), evaluation (3), whether (3), benchmarks (3), track (3), long (3), non (3), gpt (3), primer (3), supervise (3), supervision (3), 2301 (3), 01751 (3), transparent (3), oecd (3), verification (3), detecting (3), summaries (3), revolution (3), talks (3), essays (2), github (2), bher (2), webppl (2), earlier (2), worked (2), helps (2), analysts (2), type (2), bachelor (2), school (2), 2009 (2), explained (2), derive (2), compute (2), infer (2), jbt (2), cogsci2010 (2), 2010 (2), richly (2), relations (2), subparts (2), thought (2), sets (2), true (2), heuristics (2), aistats2011 (2), 14th (2), transformational (2), compilation (2), arbitrary (2), traces (2), compilers (2), infrastructure (2), etc (2), minimal (2), overhead (2), original (2), nips2011 (2), jeffrey (2), siskind (2), allow (2), analysis (2), examine (2), purpose (2), blocks (2), construct (2), global (2), irvin (2), hwang (2), inducing (2), outlines (2), express (2), extension (2), generalization (2), improves (2), probability (2), transformations (2), types (2), transformation (2), common (2), form (2), recursion (2), mccoy (2), tobias (2), gerstenberg (2), bob (2), intervene (2), situation (2), scenarios (2), empirical (2), rather (2), attributions (2), others (2), speakers (2), utterances (2), optimally (2), interpret (2), utterance (2), predicts (2), listener (2), fit (2), statistical (2), relational (2), uai (2), starai (2), 1206 (2), 3555 (2), computing (2), discrete (2), turns (2), build (2), potentially (2), point (2), during (2), local (2), rise (2), backwards (2), likely (2), poster (2), cogsys2013 (2), mind (2), wide (2), linguistics (2), easy (2), discuss (2), algorithmic (2), address (2), topics2013 (2), topics (2), nips2013 (2), jessica (2), taylor (2), estimate (2), quickly (2), parameter (2), cogsci2014 (2), sixth (2), acquire (2), continue (2), enrich (2), expand (2), exposed (2), varied (2), across (2), contrast (2), acquisition (2), suitable (2), forces (2), govern (2), interact (2), unfolding (2), learners (2), inferring (2), novel (2), laws (2), microworlds (2), movies (2), generally (2), errors (2), approximate (2), dippl (2), note (2), accessed (2), formal (2), implementing (2), javascript (2), several (2), particle (2), filtering (2), ideas (2), semantic (2), procedural (2), massachusetts (2), productive (2), metaphor (2), half (2), power (2), domains (2), naturally (2), expressed (2), initial (2), necessary (2), efficiently (2), tractable (2), third (2), generic (2), appropriate (2), finally (2), degen (2), judith (2), seventh (2), ognitive (2), cience (2), ociety (2), provoke (2), sophisticated (2), behavior (2), inferred (2), components (2), needed (2), account (2), siddharth (2), sequential (2), sequences (2), highly (2), same (2), depth (2), hidden (2), preliminary (2), ritchie (2), callsite (2), 19th (2), aistats (2), complete (2), proposal (2), core (2), function (2), reduces (2), runtime (2), cases (2), ignorant (2), 30th (2), important (2), hyperbolic (2), discounting (2), incorporate (2), who (2), subjects (2), take (2), abel (2), protocols (2), teacher (2), tailored (2), particular (2), schema (2), conduct (2), protocol (2), agentmodels (2), filan (2), mdps (2), motivation (2), create (2), plan (2), simulating (2), multi (2), games (2), directly (2), experience (2), otherwise (2), contained (2), cannot (2), vague (2), care (2), background (2), through (2), follow (2), causes (2), mechanism (2), quality (2), conversations (2), pay (2), helpful (2), contributions (2), market (2), needs (2), incentives (2), depends (2), outline (2), addressing (2), project (2), applications (2), girish (2), sastry (2), proc (2), 17th (2), autonomous (2), multiagent (2), aamas (2), error (2), single (2), free (2), having (2), prevent (2), labor (2), learner (2), deep (2), fails (2), catastrophic (2), less (2), successful (2), does (2), found (2), scale (2), approximations (2), resources (2), approximation (2), bottom (2), ishita (2), dasgupta (2), demi (2), guo (2), samuel (2), gershman (2), 40th (2), frontier (2), serve (2), problems (2), detection (2), focus (2), pairs (2), word (2), driven (2), snli (2), out (2), chris (2), cundy (2), ryan (2), carey (2), kenton (2), thomas (2), mcgrath (2), andrew (2), schreiber (2), include (2), medical (2), judging (2), judgment (2), partly (2), prediction (2), designed (2), estimation (2), miller (2), context (2), claims (2), lot (2), progress (2), neel (2), alex (2), eli (2), lifland (2), lewis (2), tunstall (2), abhishek (2), thakur (2), pegah (2), maham (2), jess (2), riedel (2), emmie (2), hine (2), carolyn (2), ashurst (2), paul (2), sedille (2), alexis (2), carlier (2), michael (2), noetel (2), neurips (2), promise (2), don (2), baselines (2), sometimes (2), average (2), benefits (2), post (2), term (2), capabilities (2), hard (2), maggie (2), appleton (2), supervising (2), latent (2), need (2), refining (2), accuracy (2), automated (2), keep (2), opportunities (2), publishing (2), scientific (2), literature (2), 2310 (2), 10627 (2), summarizing (2), chatgpt (2), claude (2), stanford (2), 2024 (2), wise (2), writing (2), rapidly (2), epistemics (2), growing (2), steer (2), truth (2), seeking (2), email, professional, profile, linkedin, scholar, elsewhere, exact, cosh, repository, forest, hiring, mastersthesis, osnabrueck2009, osnabrueck, bscthesis, complexly, center, candidate, hypothesis, plausible, consisting, generating, feasible, let, likelihood, combination, quantitatively, reliably, demonstration, osnabrück, car, house, simply, collection, features, attempts, builds, viewing, rule, inductively, modeled, described, exploratory, investigate, vary, prototypes, recursions, categorization, exemplar, 32nd, general, straightforward, engines, named, position, trace, names, conjunction, database, holding, variables, encode, naming, reuse, profilers, implying, low, technique, imperative, compiled, version, eliminates, interpretive, matlab, jms, modelers, specify, syntax, resembles, modern, readable, format, compiler, emph, craft, gradients, monad, side, executing, easily, coded, operator, overloading, automatic, differentiation, gradient, provenance, tracking, construction, arxiv2011, allows, choosing, algebraic, begin, generates, greater, hence, incrementally, extent, motivated, explanatory, abstractions, merges, subexpressions, anti, unification, deargumentation, simplifies, finds, key, lists, parameterized, jmc, counterfactuals, causal, proportion, improved, raises, alternatives, intervening, event, series, coin, tossing, proposed, sample, deterministically, switching, differently, own, compared, believe, prize, arxiv2012, takes, interpreter, marginalizer, impossible, presence, graph, sum, product, network, cyclic, subproblems, explicit, corresponds, fixed, iteration, topological, nipsprobprog2012, nipsworkshop2012, accumulates, compose, overall, targeted, expression, stochastically, giving, parse, means, blocked, issn, 1389, 0417, 1016, cogsys, 003, traditionally, viewed, operation, explicitly, part, desires, intentions, captures, flexibility, inherent, directions, fields, posed, kinds, gibbs, invest, upfront, rapid, online, queries, joint, root, nodes, accumulate, conditional, constitute, leading, observation, explanation, estimated, converge, asymptotically, convergence, block, efficiency, regimes, bayes, nets, theories, timescales, kemp, griffiths, behave, suggestive, complemented, bottomup, best, unify, description, uncertain, seen, communities, embedding, host, designing, small, ppl, embedded, universal, priority, enumeration, expose, stack, addresses, drawn, parsing, pragmatics, graphics, phdthesis, astthesis, month, develops, respect, mental, manipulation, representational, experimental, match, proceed, fact, unlike, formalisms, insight, dual, intent, accounts, compiling, reusing, subcomputations, lets, leverage, compile, recognition, speed, spaces, substantiate, claim, outlining, opened, investigation, hawkins2015why, uller, useful, paper, formulate, family, pieces, classic, effect, psycholinguistics, informativeness, varies, keeping, jointly, among, pragmatic, cues, select, stuhlmueller2015coarse, smc, filter, practical, sequence, interpolate, intractable, usually, geometric, averages, themselves, annealing, sensitive, generate, increasing, detail, ising, disparity, variation, factorial, evans2015learning, nipsworkshop2015, ritchie2015c3, 02151, popular, simplicity, deterministic, execute, runtimes, requiring, every, continuation, passing, style, cps, calls, 100x, complexity, linear, size, nearly, magnitude, speedup, application, evans2016learning, aaai2016, posts, photos, user, jobs, activities, rewarding, past, inform, likes, according, treat, observed, deviate, temporally, subject, inconsistency, behavioral, abel2016agent, nips2016, dramatically, various, aspects, developed, enable, guidance, underlying, offering, effective, specialized, procedures, beneficial, strong, assumptions, inner, workings, action, pruning, shaping, simulation, effectiveness, mdp, pomdp, describes, implements, richer, rationality, implemented, differentiable, recursively, selves, opponents, run, browser, modify, extend, assumes, includes, introductions, pomdps, myopic, dialogmarkets, frequently, career, right, boyfriend, money, capital, turkey, though, tend, particulars, doctor, stomach, pain, won, give, clarifying, start, intermittent, notes, creating, resolve, starts, individual, asks, pledges, contribute, responses, distribute, incentivize, valuable, delegate, contribution, gives, grounding, meta, turn, ensure, dialogs, sufficiently, off, ground, name, potential, saunders2017trial, 1707, 05173, involve, dangerous, haven, actions, cause, harm, mistake, harms, damage, ready, currently, reduce, imitate, atari, being, overseen, four, affecting, whereas, forgetting, eliminate, adversarial, extrapolating, infeasible, amount, extensions, train, catastrophe, ullman2018learning, cogpsy2017, estimating, simultaneously, interacting, consistent, indicative, solving, computationally, demanding, limited, complement, relies, combines, strengths, taking, departure, dasgupta2018evaluating, 1802, 04302, quest, semantics, infinite, expressions, finite, vocabulary, attempted, vector, sentences, input, paraphrase, sentiment, here, critical, capacity, assigned, categories, entailment, contradiction, neutral, solved, degree, art, conneau, poor, indicating, fail, analyze, rules, largely, ecologically, augmenting, held, loss, highlights, techreport, evans2018predicting, institution, humanity, oxford, tr2018, effortlessly, realworld, extended, analytical, gathering, external, discussing, collaborators, criminal, personalized, recommendations, rich, content, books, there, great, demand, unreliable, intrinsically, assembling, expensive, label, collecting, predict, spend, collect, after, seconds, minutes, combined, smaller, quantity, resulting, related, semi, collaborative, testing, fermi, back, envelope, veracity, political, statements, collected, 000, 800, define, contains, tested, unfortunately, limitations, testbed, mistakes, available, oughtinc, psj, fhi, ought2020arguments, january, studying, conditions, group, accomplish, focused, dividing, created, movie, reviews, tried, review, identifying, implausible, participants, disagreed, were, therefore, brittle, diverse, besides, quotes, larger, trees, guidelines, update, findings, alex2021raft, 2109, 14076, pre, completing, specific, soon, far, reserved, assistants, measure, settings, annotated, focuses, occurring, setup, mirrors, deployment, evaluations, reveal, areas, struggle, texts, classes, reflecting, expertise, scores, exceed, leaderboard, improvements, translate, stuhlmueller2022primer, urldate, stuhlmueller2022process, spectrum, understandable, decompositions, optimization, final, devoted, differential, access, forecasting, policy, theoretical, gaming, measures, thus, attractors, varying, degrees, once, architecture, entrenched, move, away, lock, applies, powerful, primarily, air, crucial, push, toward, now, reppert2023iterated, debugging, lms, either, compositionally, composition, offers, interpretability, safety, remain, competitive, developing, compositions, zooming, failing, ice, tool, visualizing, describing, placebo, randomized, controlled, adherence, nlp, qasper, studies, interpretable, incollection, byun2023elicit, publisher, paris, 1787, a8d820bd, www, science_a8d820bd, full, tools_fec8a6ab, html, next, decade, summarisation, essay, assistant, summarise, george2023factored, plagues, llms, bad, really, abstractive, sota, summarization, halueval, achieving, hallucinate, 16k, summary, correct, critiques, lowers, subtle, advise, caution, published, extraction, wiesp, ijcnlp, aacl, seminar, ended, futurati, podcast, labs, curve, 2025, bio, living, radically, podcasts, going, pharma, unsolved, was, expressing, exploring, minds, interested, mean, still, informs, reasons, foundations, advancing, urgent, faster, produces, output, looks, years, decades, transformative, rare, matters, governments, allocate, intuition, companies, strategic, overconfidence, misleading, irreproducible, figuring, especially, billions, dollars, millions, lives, read, cofounded, profit, explored, scalable, alignment, phd, josh, cofounder, ceo, mission, today, sciences,
Text of the page (random words):
trial author william saunders and girish sastry and andreas stuhlm u ller and owain evans booktitle proc of the 17th international conference on autonomous agents and multiagent systems aamas 2018 title trial without error towards safe reinforcement learning via human intervention year 2018 url https arxiv org abs 1707 05173 dialog markets andreas stuhlmüller 2016 technical report abstract people frequently ask google questions like what career is right for me how can i find a boyfriend or how can i make more money while google can answer self contained questions like what is the capital of turkey it cannot answer vague questions even though these tend to be the questions people care about the most answering these questions requires background information about the person who asked good advice is tailored to the particulars of the asker s situation a natural way to elicit such background is through a dialog in which follow up questions are asked if you ask a doctor why do i have stomach pain they won t just give you an answer they will ask clarifying questions when did it start where is it is it constant or intermittent and only then they suggest possible causes in this set of notes i propose dialog markets a mechanism for creating high quality conversations that resolve vague questions a dialog starts when an individual asks a question and pledges a reward that will be used to pay for helpful contributions humans and machines then contribute follow up questions and other responses the core problem that the market mechanism needs to solve is how to distribute reward over contributions such that we incentivize conversations that are as valuable as possible for the asker we can view this as a particular set of questions that we can delegate to the market how much should the asker pay for this contribution this view gives rise to the grounding problem for dialog markets incentives will be aligned if answers to this meta level question are good but that in turn depends on aligned incentives i outline some approaches towards addressing this problem finally i suggest an implementation plan i discuss how to ensure that early dialogs have sufficiently high quality to get the project off the ground and name a few potential initial applications bibtex misc dialogmarkets author andreas stuhlm u ller title dialog markets year 2016 url https stuhlmueller org dialog markets modeling agents with probabilistic programs owain evans andreas stuhlmüller john salvatier and daniel filan 2017 abstract this book describes and implements models of rational agents for po mdps and reinforcement learning one motivation is to create richer models of human planning which capture human biases and bounded rationality agents are implemented as differentiable functional programs in a probabilistic programming language based on javascript agents plan by recursively simulating their future selves or by simulating their opponents in multi agent games our agents and environments run directly in the browser and are easy to modify and extend the book assumes basic programming experience but is otherwise self contained it includes short introductions to planning as inference mdps pomdps inverse reinforcement learning hyperbolic discounting myopic planning and multi agent planning bibtex misc agentmodels author owain evans and andreas stuhlm u ller and john salvatier and daniel filan title modeling agents with probabilistic programs year 2017 keywords agents reinforcement learning mdp pomdp interactive web book url https agentmodels org agent agnostic human in the loop reinforcement learning david abel john salvatier andreas stuhlmüller and owain evans 2016 future of interactive learning machines workshop at nips 2016 abstract providing reinforcement learning agents with expert advice can dramatically improve various aspects of learning to this end prior work has developed teaching protocols that enable agents to learn efficiently in complex environments in many of these methods the teacher s guidance is tailored to agents with a particular representation or underlying learning scheme offering effective but highly specialized teaching procedures in this work we introduce protocol programs an agent agnostic schema for human in the loop reinforcement learning our goal is to incorporate the beneficial properties of a human teacher into reinforcement learning without making strong assumptions about the inner workings of the agent we show how to represent existing approaches such as action pruning reward shaping and training in simulation as special cases of our schema and conduct preliminary experiments evaluating the effectiveness of protocols in simple domains bibtex inproceedings abel2016agent author david abel and john salvatier and andreas stuhlm u ller and owain evans title agent agnostic human in the loop reinforcement learning booktitle future of interactive learning machines workshop at nips 2016 year 2016 keywords reinforcement learning interactive human in the loop url https stuhlmueller org papers human rl nips2016 pdf learning the preferences of ignorant inconsistent agents owain evans andreas stuhlmüller and noah d goodman 2016 proceedings of the 30th aaai conference on artificial intelligence aaai 2016 abstract an important use of machine learning is to learn what people value what posts or photos should a user be shown which jobs or activities would a person find rewarding in each case observations of people s past choices can inform our inferences about their likes and preferences if we assume that choices are approximately optimal according to some utility function we can treat preference inference as bayesian inverse planning that is given a prior on utility functions and some observed choices we invert an optimal decision making process to infer a posterior distribution on utility functions however people often deviate from approximate optimality they have false beliefs their planning is sub optimal and their choices may be temporally inconsistent due to hyperbolic discounting and other biases we demonstrate how to incorporate these deviations into algorithms for preference inference by constructing generative models of planning for agents who are subject to false beliefs and time inconsistency we explore the inferences these models make about preferences beliefs and biases we present a behavioral experiment in which human subjects perform preference inference given the same observations of choices as our model results show that human subjects like our model explain choices in terms of systematic deviations from optimal behavior and suggest that they take such deviations into account when inferring preferences bibtex inproceedings evans2016learning author owain evans and andreas stuhlm u ller and noah d goodman title learning the preferences of ignorant inconsistent agents booktitle proceedings of the 30th aaai conference on artificial intelligence aaai 2016 year 2016 keywords preference inference bayesian learning cognitive biases url https stuhlmueller org papers preferences aaai2016 pdf c3 lightweight incrementalized mcmc for probabilistic programs using continuations and callsite caching daniel ritchie andreas stuhlmüller and noah d goodman 2016 proceedings of the 19th international conference on artificial intelligence and statistics aistats abstract lightweight source to source transformation approaches to implementing mcmc for probabilistic programming languages are popular for their simplicity support of existing deterministic code and ability to execute on existing fast runtimes however they are also slow requiring a complete re execution of the program on every metropolis hastings proposal we present a new extension to the lightweight approach c3 which enables efficient incrementalized re execution of mh proposals c3 is based on two core ideas transforming probabilistic programs into continuation passing style cps and caching the results of function calls we show that on several common models c3 reduces proposal runtime by 20 100x in some cases reducing runtime complexity from linear in model size to constant we also demonstrate nearly an order of magnitude speedup on a complex inverse procedural modeling application bibtex inproceedings ritchie2015c3 author daniel ritchie and andreas stuhlm u ller and noah d goodman title c3 lightweight incrementalized mcmc for probabilistic programs using continuations and callsite caching booktitle proceedings of the 19th international conference on artificial intelligence and statistics aistats year 2016 keywords probabilistic programming inference mcmc metropolis hastings url https arxiv org abs 1509 02151 learning the preferences of bounded agents owain evans andreas stuhlmüller and noah d goodman 2015 poster at the nips 2015 workshop on bounded optimality bibtex inproceedings evans2015learning author owain evans and andreas stuhlm u ller and noah d goodman booktitle nips 2015 workshop on bounded optimality title learning the preferences of bounded agents year 2015 url https stuhlmueller org papers preferences nipsworkshop2015 pdf coarse to fine sequential monte carlo for probabilistic programs andreas stuhlmüller robert x d hawkins n siddharth and noah d goodman 2015 technical report arxiv 1509 02962 cs ai abstract many practical techniques for probabilistic inference require a sequence of distributions that interpolate between a tractable distribution and an intractable distribution of interest usually the sequences used are simple e g based on geometric averages between distributions when models are expressed as probabilistic programs the models themselves are highly structured objects that can be used to derive annealing sequences that are more sensitive to domain structure we propose an algorithm for transforming probabilistic programs to coarse to fine programs which have the same marginal distribution as the original programs but generate the data at increasing levels of detail from coarse to fine we apply this algorithm to an ising model its depth from disparity variation and a factorial hidden markov model we show preliminary evidence that the use of coarse to fine models can make existing generic inference algorithms more efficient bibtex article stuhlmueller2015coarse author andreas stuhlm u ller and robert x d hawkins and n siddharth and noah d goodman archiveprefix arxiv eprint 1509 02962 primaryclass cs ai title coarse to fine sequential monte carlo for probabilistic programs year 2015 keywords probabilistic programming inference smc particle filter coarse to fine url https arxiv org abs 1509 02962 why do you ask good questions provoke informative answers robert x d hawkins andreas stuhlmüller judith degen and noah d goodman 2015 proceedings of the thirty seventh annual conference of the cognitive science society abstract what makes a question useful what makes an answer appropriate in this paper we formulate a family of increasingly sophisticated models of question answer behavior within the rational speech act framework we compare these models based on three different pieces of evidence first we demonstrate how our answerer models capture a classic effect in psycholinguistics showing that an answerer s level of informativeness varies with the inferred questioner goal while keeping the question constant second we jointly test the questioner and answerer components of our model based on empirical evidence from a question answer reasoning game third we examine a special case of this game to further distinguish among the questioner models we find that sophisticated pragmatic reasoning is needed to account for some of the data people can use questions to provide cues to the answerer about their interest and can select answers that are informative about inferred interests bibtex inproceedings hawkins2015why author hawkins robert x d and stuhlm uller andreas and degen judith and goodman noah d booktitle proceedings of the thirty seventh annual conference of the c ognitive s cience s ociety title why do you ask good questions provoke informative answers year 2015 modeling cognition with probabilistic programs representations and algorithms andreas stuhlmüller 2015 ph d thesis massachusetts institute of technology abstract this thesis develops probabilistic programming as a productive metaphor for understanding cognition both with respect to mental representations and the manipulation of such representations in the first half of the thesis i demonstrate the representational power of probabilistic programs in the domains of concept learning and social reasoning i provide examples of richly structured concepts defined in terms of systems of relations subparts and recursive embeddings that are naturally expressed as programs and show initial experimental evidence that they match human generalization patterns i then proceed to models of reasoning about reasoning a domain where the expressive power of probabilistic programs is necessary to formalize our intuitive domain understanding due to the fact that unlike previous formalisms probabilistic programs allow conditioning to be represented a model not just applied to a model i illustrate this insight with programs that model nested reasoning in game theory artificial intelligence and linguistics in the second half i develop three inference algorithms with the dual intent of showing how to efficiently compute the marginal distributions defined by probabilistic programs and providing building blocks for process level accounts of human cognition first i describe a dynamic programming algorithm for computing the marginal distribution of discrete probabilistic programs by compiling to systems of equations and show that it can make inference in models of reasoning about reasoning tractable by merging and reusing subcomputations second i introduce the setting of amortized inference and show how learning inverse models lets us leverage samples generated by other inference algorithms to compile probabilistic models into fast recognition functions third i develop a generic approach to coarse to fine inference in probabilistic programs and provide evidence that it can speed up inference in models with large state spaces that have appropriate hierarchical structure finally i substantiate the claim that probabilistic programming is a productive metaphor by outlining new research questions that have been opened up by this line of investigation bibtex phdthesis astthesis author andreas stuhlm u ller title modeling cognition with probabilistic programs representations and algorithms school massachusetts institute of technology year 2015 month 6 the design and implementation of probabilistic programming languages noah d goodman and andreas stuhlmüller 2014 abstract probabilistic programming languages ppls unify techniques for the formal description of computation and for the representation and use of uncertain knowledge ppls have seen recent interest ...
|
