Blade Runner, Directed By, Ridley Scott] [Blade Runner, Written By, Philip K

Total Page:16

File Type:pdf, Size:1020Kb

Blade Runner, Directed By, Ridley Scott] [Blade Runner, Written By, Philip K Teaching Machines to Understand Natural Language Antoine Bordes Cap’18 – Rouen – June 19th, 2018 Mission “Advancing the state of the art of AI through open research for the benefit of all” Values • Openness: publish and open-source • Freedom: researchers have complete control on their agenda • Collaboration: with internal and external partners • Excellence: focus on most impactful projects (publish or perish) • Scale: operate at large scale (computation, data) 3 Talking to Machines Language should be the most natural interface between the digital and the physical world Digital personal assistants: Grant accessibility for people with disabilities “What is in this picture?” Break loneliness “I’m not feeling well..” Mediate and monitor our online presence “Did anybody talk about me or my family lately?” “Why do I see this content?” Machines Understanding Language Current systems are still far from natural interactions. Intelligent machines should be able to: 1. Build (causal) models of the world as internal knowledge representations; 2. Ground language into them to interpret it; 3. Leverage compositionality to learn quickly and to update them. Machines are missing the ability to learn high-level abstractions “Building machines that learn and think like people” Lake et al. Behavioral and Brain Sciences ’17 “Generalization without systematicity: On the com- positional skills of sequence-to-sequence recurrent networks” Lake & Baroni. Arxiv’18 Two Paradigms Neural Networks Symbolic Systems Convolutional nets, Recurrent nets, etc. Ontologies, KBs, Inductive Logic Programming, etc. • Scale to very large datasets • Can be used by non domain • Small scale conditions experts • Require heavy expert knowledge • Robust to noise and ambiguity in • Very brittle with noisy, ambiguous data data • Game changers in multiple • Limited applicative success applications • Learn from few samples • Very data hungry (mostly • Can rely on high-level abstractions supervised data) • Interpretable, provable • Can not learn easily new tasks • Complex reasoning from old ones • Not interpretable • Relatively simple reasoning Conjecture Neural networks (and friends) can be adapted to be the best of both worlds Architectures at the interplay between symbolic and continuous models encode structure such as symbolic databases discover structure in (unstructured) data operate and are trained with symbolic components (even programs) Training will have to overcome new challenges Optimization mixing differentiable and non-differentiable parts Reinforcement learning requires rewards and environment This talk: Question Answering This talk focus on NNs learning to use encyclopedic knowledge Task: question answering (Q&A) because of ease of evaluation 1. How neural networks can be adapted to reason over symbols? 2. How vector spaces can encode structured data from knowledge bases? Neural Networks + Symbolic Memories bAbI Tasks • Set of 20 tasks testing basic reasoning capabilities for QA from stories • Short stories are generated from a simulation • Easy to interpret results / test a broad range of properties John dropped the milk. The suitcase is bigger than the chest. John took the milk there. The box is bigger than the chocolate. Sandra went to the bathroom. The chest is bigger than the chocolate. John moved to the hallway. The chest fits inside the container. Mary went to the bedroom. The chest fits inside the box. Where is the milk ? Hallway Does the suitcase fit in the chocolate? no Task 3: Two supporting facts Task 18: Size reasoning • Useful to foster innovation: cited +300 times “Towards AI-complete question answering: A set of prerequisite toy tasks.” Weston et al. ICLR’16 bAbI Tasks John dropped the milk. The suitcase is bigger than the chest. John took the milk there. The box is bigger than the chocolate. Sandra went to the bathroom. The chest is bigger than the chocolate. John moved to the hallway. The chest fits inside the container. Mary went to the bedroom. The chest fits inside the box. Where is the milk ? Hallway Does the suitcase fit in the chocolate? no Task 3: Two supporting facts Task 18: Size reasoning Very challenging for standard neural networks: • LSTMs: 63% accuracy and 4 tasks completed with 10k training examples “Towards AI-complete question answering: A set of prerequisite toy tasks.” Weston et al. ICLR’16 Memory Networks Models that combine symbolic memory with learning component that can read it. Memory Networks: Learn to reason with attention over their memory Can be trained end-to-end without supervising how the memory is used using cross-entropy or ranking loss Multiple related models appeared around the same time and since then: Stack-RNN (Joulin et Mikolov, ‘15), RNNSearch (Bahdanau et al., ‘15) Neural Turin Machines (Graves et al. ‘15) , etc. “Memory Networks” Weston et al. ICLR’15 / “End-to-end Memory Networks.” Sukhbaatar et al. NIPS’1S Question + memory John hallway milk hallway office Memory Networks on bAbI … m5: Mary went back to the bedroom. , …) m4 : John move , m d to the hallway. , m 3 m32 : Sandra went to the bathr , m oom. backpropagation(m 1 End-to-endm2: Jtrainingohn tMemoriesook tthroughhe milk t here. m1: John dropped the milk. Where is the milk? John dropped the milk. John took the milk there. Sandra went to the bathroom. John moved to the hallway. Mary went back to the bedroom. Attention during memory lookups Story (1: 1 supporting fact) Support Hop 1 Hop 2 Hop 3 Story (2: 2 supporting facts) Support Hop 1 Hop 2 Hop 3 Daniel went to the bathroom. 0.00 0.00 0.03 J ohn dropped the milk. 0.06 0.00 0.00 Mary travelled to the hallway. 0.00 0.00 0.00 J ohn took the milk there. yes 0.88 1.00 0.00 J ohn went to the bedroom. 0.37 0.02 0.00 Sandra went back to the bathroom. 0.00 0.00 0.00 J ohn travelled to the bathroom. yes 0.60 0.98 0.96 J ohn moved to the hallway. yes 0.00 0.00 1.00 Mary went to the office. 0.01 0.00 0.00 Mary went back to the bedroom. 0.00 0.00 0.00 Where is J ohn? Answer: bathroom Prediction: bathroom Where is the milk? Answer: hallway Prediction: hallway Story (16: basic induction) Support Hop 1 Hop 2 Hop 3 Story (18: size reasoning) Support Hop 1 Hop 2 Hop 3 Brian is a frog. yes 0.00 0.98 0.00 The suitcase is bigger than the chest. yes 0.00 0.88 0.00 Lily is gray. 0.07 0.00 0.00 The box is bigger than the chocolate. 0.04 0.05 0.10 Brian is yellow. yes 0.07 0.00 1.00 The chest is bigger than the chocolate. yes 0.17 0.07 0.90 J ulius is green. 0.06 0.00 0.00 The chest fits inside the container. 0.00 0.00 0.00 Greg is a frog. yes 0.76 0.02 0.00 The chest fits inside the box. 0.00 0.00 0.00 What color is GregHow? Answer: yellow Paboutrediction: yellow onDo esreal the suitcase fit in languagethe chocolate? Answer: no Predictio n: no Figure 2: Example predictions on the QAdata?tasks of [21]. We show the labeled supporting facts • LSTMs:(support) from the dataset which MemN2N does not use during63%training, accuracyand the probabilitiesand 4 tasksp of completed each hop used by themodel during inference. MemN2N successfully learns to focus on thecorrect (supportingwith 10ksentences. ex) • Memory Networks (3 hops): 96% accuracy and 17 tasks completed (with 10k ex) Penn Treebank Text8 #of #of memory Valid. Test #of #of memory Valid. Test “End-to-end• Recently ModelMemory Networks.” (2017),hidden hopsSukhbaatar EntNetssize et al. perp.and NIPS’1S Query-reductionperp. hidden hops size Networksperp. perp. solve all 20 tasks. RNN [15] 300 - - 133 129 500 - - - 184 LSTM [15] 100 - - 120 115 500 - - 122 154 SCRN [15] 100 - - 120 115 500 - - - 161 MemN2N 150 2 100 128 121 500 2 100 152 187 150 3 100 129 122 500 3 100 142 178 150 4 100 127 120 500 4 100 129 162 150 5 100 127 118 500 5 100 123 154 150 6 100 122 115 500 6 100 124 155 150 7 100 120 114 500 7 100 118 147 150 6 25 125 118 500 6 25 131 163 150 6 50 121 114 500 6 50 132 166 150 6 75 122 114 500 6 75 126 158 150 6 100 122 115 500 6 100 124 155 150 6 125 120 112 500 6 125 125 157 150 6 150 121 114 500 6 150 123 154 150 7 200 118 111 -- - - - Table 2: The perplexity on the test sets of Penn Treebank and Text8 corpora. Note that increasing thenumber of memory hopsimproves performance. Figure 3: Average activation weight of memory positions during 6 memory hops. White color indicates where the model is attending during the kth hop. For clarity, each row is normalized to havemaximumvalueof 1. A model istrained on(left) Penn Treebank and (right) Text8 dataset. 5 LanguageModeling Experiments The goal in language modeling is to predict the next word in a text sequence given the previous wordsx. Wenow explain how our model can easily beapplied to this task. Wenow operate on word level, as opposed to thesentence level. Thus theprevious N words in the sequence (including the current) are embedded into memory separately. Each memory cell holds only asingle word, so there is no need for the BoW or linear mapping representations used in the QA tasks. Weemploy thetemporal embedding approach of Section 4.1. Since there is no longer any question, q in Fig. 1 is fixed to a constant vector 0.1 (without embedding). The output softmax predicts which word in the vocabulary (of size V ) is next in the sequence.
Recommended publications
  • Goodbye Cinema, Hello Cinephilia Other Books by Jonathan Rosenbaum
    Goodbye Cinema, Hello Cinephilia Other Books by Jonathan Rosenbaum Rivette: Texts and Interviews (editor, 1977) Orson Welles: A Critical View, by André Bazin (editor and translator, 1978) Moving Places: A Life in the Movies (1980) Film: The Front Line 1983 (1983) Midnight Movies (with J. Hoberman, 1983) Greed (1991) This Is Orson Welles, by Orson Welles and Peter Bogdanovich (editor, 1992) Placing Movies: The Practice of Film Criticism (1995) Movies as Politics (1997) Another Kind of Independence: Joe Dante and the Roger Corman Class of 1970 (coedited with Bill Krohn, 1999) Dead Man (2000) Movie Wars: How Hollywood and the Media Limit What Films We Can See (2000) Abbas Kiarostami (with Mehrmax Saeed-Vafa, 2003) Movie Mutations: The Changing Face of World Cinephilia (coedited with Adrian Martin, 2003) Essential Cinema: On the Necessity of Film Canons (2004) Discovering Orson Welles (2007) The Unquiet American: Trangressive Comedies from the U.S. (2009) Goodbye Cinema, Hello Cinephilia Film Culture in Transition Jonathan Rosenbaum the university of chicago press | chicago and london Jonathan Rosenbaum wrote for many periodicals (including the Village Voice, Sight and Sound, Film Quarterly, and Film Comment) before becoming principal fi lm critic for the Chicago Reader in 1987. Since his retirement from that position in March 2008, he has maintained his own Web site and continued to write for both print and online publications. His many books include four major collections of essays: Placing Movies (California 1995), Movies as Politics (California 1997), Movie Wars (a cappella 2000), and Essential Cinema (Johns Hopkins 2004). The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2010 by The University of Chicago All rights reserved.
    [Show full text]
  • Key-Value Memory Networks for Directly Reading Documents
    Key-Value Memory Networks for Directly Reading Documents Alexander H. Miller1 Adam Fisch1 Jesse Dodge1,2 Amir-Hossein Karimi1 Antoine Bordes1 Jason Weston1 1Facebook AI Research, 770 Broadway, New York, NY, USA 2Language Technologies Institute, Carnegie Mellon University, Pittsburgh, PA, USA {ahm,afisch,jessedodge,ahkarimi,abordes,jase}@fb.com Abstract can be used to query such databases (Berant et al., 2013; Kwiatkowski et al., 2013; Fader et al., 2014). Directly reading documents and being able to Unfortunately, KBs have intrinsic limitations such answer questions from them is an unsolved as their inevitable incompleteness and fixed schemas challenge. To avoid its inherent difficulty, ques- that cannot support all varieties of answers. Since tion answering (QA) has been directed towards information extraction (IE) (Craven et al., 2000), in- using Knowledge Bases (KBs) instead, which has proven effective. Unfortunately KBs often tended to fill in missing information in KBs, is neither suffer from being too restrictive, as the schema accurate nor reliable enough, collections of raw tex- cannot support certain types of answers, and tual resources and documents such as Wikipedia will too sparse, e.g. Wikipedia contains much more always contain more information. As a result, even if information than Freebase. In this work we KBs can be satisfactory for closed-domain problems, introduce a new method, Key-Value Memory they are unlikely to scale up to answer general ques- Networks, that makes reading documents more tions on any topic. Starting from this observation, viable by utilizing different encodings in the ad- in this work we study the problem of answering by dressing and output stages of the memory read operation.
    [Show full text]
  • Presentazione AIXIA 2013.Key
    A Virtual Player for “Who Wants to Be a Millionaire?” based on Question Answering Piero Molino - [email protected]! Università degli Studi di Bari Aldo Moro Pierpaolo Basile, Ciro Santoro, Pasquale Lops, Marco de Gemmis and Giovanni Semeraro! Who wants to be Millionaire? € 50 € 100 € 200 50:50 € 300 € 500 € 1 000 € 2 000 € 4 000 € 8 000 Who directed Blade Runner?€ 16 000 € 32 000 € 64 000 € 125 000 A € 500 000 Harrison Ford € 1 000 000 C Philip Dick B Ridley Scott D James Cameron Methodology ✤ Obtain scored passages from a QA system! ✤ Answers are paragraphs obtained from Wikipedia or triples from DBpedia! ✤ Calculate the confidence in each of the four answers using this information! ✤ Choose the answer according to the confidence Who directed Blade Runner? Article Title Passage Text Score Sir Ridley Scott (born 30 November 1937) is an English film director and producer. Following Ridley Scott his commercial breakthrough with Alien (1979), his best-known works are the sci-fi classic Blade Runner (1982) and the best picture Oscar-winner Gladiator (2000). 5.3 Blade Runner is a 1982 American dystopian science fic- tion action film directed by Ridley Scott and starring Harrison Ford, Rutger Hauer, and Sean Young. The screenplay, written by Blade Runner Hampton Fancher and David Peo- ples, is loosely based on the novel Do Androids Dream of 5.1 Electric Sheep? by Philip K. Dick. Director Ridley Scott and the film’s producers ”spent months” meeting and discussing the Blade Runner role with Dustin Hoffman, who eventually departed over differences in vision.
    [Show full text]
  • Film, Philosophy Andreligion
    FILM, PHILOSOPHY AND RELIGION Edited by William H. U. Anderson Concordia University of Edmonton Alberta, Canada Series in Philosophy of Religion Copyright © 2022 by the authors. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Vernon Art and Science Inc. www.vernonpress.com In the Americas: In the rest of the world: Vernon Press Vernon Press 1000 N West Street, Suite 1200, C/Sancti Espiritu 17, Wilmington, Delaware 19801 Malaga, 29006 United States Spain Series in Philosophy of Religion Library of Congress Control Number: 2021942573 ISBN: 978-1-64889-292-9 Product and company names mentioned in this work are the trademarks of their respective owners. While every care has been taken in preparing this work, neither the authors nor Vernon Art and Science Inc. may be held responsible for any loss or damage caused or alleged to be caused directly or indirectly by the information contained in it. Every effort has been made to trace all copyright holders, but if any have been inadvertently overlooked the publisher will be pleased to include any necessary credits in any subsequent reprint or edition. Cover design by Vernon Press. Cover image: "Rendered cinema fimstrip", iStock.com/gl0ck To all the students who have educated me throughout the years and are a constant source of inspiration. It’s like a splinter in your mind. ~ The Matrix Table of contents List of Contributors xi Acknowledgements xv Introduction xvii William H.
    [Show full text]
  • Blade Runner 2049 to Open the 46Th Festival Du Nouveau Cinéma
    Press release For immediate release A special presentation of Denis Villeneuve’s BLADE RUNNER 2049 Opening film of the 46th Montreal Festival du nouveau cinéma (FNC) Montreal, September 21, 2017 – Montreal Festival du nouveau cinéma (FNC) is pleased to announce that Denis Villeneuve’s latest work, BLADE RUNNER 2049, will be presented at a special screening on October 4 at Théâtre Maisonneuve de la Place des Arts as the opening film of its 46th edition, which runs from October 4 to 15, 2017. After screening his first movies, the FNC is proud to welcome the Quebec director back to his hometown with his latest opus, his fourth American feature, and one of the most highly anticipated films of the year. BLADE RUNNER 2049 is a sequel to the acclaimed classic Blade Runner, directed by Ridley Scott in 1982. In the new film, a young blade runner's discovery of a long-buried secret leads him on a quest to track down former blade runner, Rick Deckard, who has been missing for thirty years. BLADE RUNNER 2049 features a star-studded cast, led by Ryan Gosling (Drive, La La Land) and Harrison Ford (the Star Wars saga, Indiana Jones), reprising his role as Rick Deckard. Also starring in the movie, are Ana de Armas (War Dogs), Sylvia Hoeks (Renegades), Robin Wright (House of Cards, Wonder Woman), Mackenzie Davis (The Martian), Carla Juri (Brimstone), and Lennie James (The Walking Dead), with Dave Bautista (the Guardians of the Galaxy films), and Jared Leto (Dallas Buyers Club, Suicide Squad). BLADE RUNNER 2049 was produced by Andrew A.
    [Show full text]
  • HERRMANN MUSIC in HAVE GUN WILL TRAVEL and OTHER CLASSIC CBS TELEVISION SERIES by Bill Wrobel
    HERRMANN MUSIC IN HAVE GUN WILL TRAVEL AND OTHER CLASSIC CBS TELEVISION SERIES By Bill Wrobel [April 2006 1st draft but some material added June 13, 2007 for Rawhide & Perry Mason] Have you ever watched the old classic CBS series such as Have Gun Will Travel, Gunsmoke, Rawhide, Perry Mason, and Twilight Zone and ever wondered how much of Bernard Herrmann’s music you heard in various episodes? I remember when I was seven years old and watched Paladin on Saturday nights for its first season run. While I was still not cognizant at the time of whom Bernard Herrmann was by name, I recall the excitement I felt watching the show (westerns were big, so to speak, or very popular at that general period on American television). I particularly liked Paladin, and then Matt Dillon (Gunsmoke). I also associated the CBS series with pleasure because of the music heard. I did not know Herrmann’s name because no credits were given at the end of the episodes (since the music was a collection of so-called “stock music” cues from various composers edited into each episode). But even back at that young age I felt drawn to the music. Let’s focus first on the Have Gun Will Travel series since the first three seasons (as I write this paper) are now available on dvd. Now: Bernard Herrmann did indeed write an original score for the pilot episode of Have Gun Will Travel starring Richard Boone perfectly cast as Paladin (although I heard, true or not, that Randolph Scott was first offered the role).
    [Show full text]
  • Preface 1 Introduction
    Notes Preface 1. The foregoing description comes from the back cover of Paul M. Sammon’s invaluable book, Future Noir: The Making of Blade Runner (1996). In 1993 Blade Runner was selected by the Library of Congress for preservation in the United States National Film Registry as being ‘culturally, historically, or aesthetically significant.’ In 2007 it was named the second most visually influential film of all time by the Visual Effects Society. In 2012 in a poll with over 10,000 votes cast it was ranked by SFX magazine as the greatest science fiction film of all time. 2. Sammon’s comments appear in his audio commentary on the Workprint version of the film (included in the Five-Disc Ultimate Collector’s Edition). 3. Bukatman is here referring specifically to the film’s visual density, but his subsequent remarks make clear that he intends this judgment to apply to the film’s thematic elements as well. 1 Introduction 1. Although we are told in the film’s opening crawl that blade runner units are special police squads created to hunt down trespassing rep- licants, ‘blade runner’ nowhere appears in Philip K. Dick’s novel and its significance is never explained in the film. Blade Runner was sug- gested as a title for the film by Hampton Fancher, who wrote the orig- inal screenplay, after he came across it in his library as the title of a script by William Burroughs based on a novel by Alan E. Nourse about smugglers of medical supplies. Despite the absence of any thematic connection between the Nourse novel and Dick’s book, the title none- theless proved to be perfect for the film.
    [Show full text]
  • Ciudades Paralelas
    20 16 ● EDITORIAL ● CIUDADES PARALELAS. EL NEGATIVO DE LA CIUDAD / PARALLEL CITIES. THE NEGATIVE OF THE CIUDADES PARALELAS CITY. Francisco Montero–Fernández ● ARTÍCULOS ● UTOPÍA Y TRASLADO: ALPHAVILLE Y OTRAS CIUDADES EN 14 LA CIENCIA FICCIÓN DE LA DÉCADA DE 1960 / UTOPIA AND RELOCATION: ALPHAVILLE AND OTHER CITIES IN THE AÑO VII SCIENCE FICTION OF THE 1960s. Juan Antonio Cabezas–Garrido ● SENSING CITIES. ENTRE LOS LÍMITES DIFUSOS DEL ARTE Y LA ARQUITECTURA / SENSING CITIES. BETWEEN THE DIFFUSE LIMITS OF ART AND ARCHITECTURE. Esther Díaz Caro 14 ● NUEVA YORK, AL OTRO LADO DEL ESPEJO. EL CINE Y LA CIUDAD FUTURA COMO TEXTO / NEW YORK, THROUGH THE ARQUITECTURA ARQUITECTURA LOOKING–GLASS. CINEMA AND THE FUTURE CITY AS A TEXT. Luis Miguel Lus Arana ● DERIVAS URBANAS Y CARTOGRAFÍAS LITERARIAS. RETRATOS PATRIMONIALES DEL PRESENTE / URBAN DÉRIVES AND LITERARY CARTOGRAPHIES: HERITAGE PORTRAITS OF THE PRESENT. Joaquín Carlos Ortiz de Villajos Carrera ● PAMPLONA OTRA: EL ESCENARIO DE LOS ENCUENTROS 72 / A DIFFERENT PAMPLONA: THE VENUES OF THE ENCUENTROS 72 ART FESTIVAL. Marta García Alonso ● LA CONSTRUCCIÓN DE UN BARRIO MODERNO VISTA DESDE SUS ESPACIOS LIBRES: HUERTA DEL REY (VALLADOLID) / THE BUILDING OF PROGRESO A MODERN DISTRICT SEEN FROM THE PERSPECTIVE OF ITS OPEN SPACES: HUERTA DEL REY (VALLADOLID). Marina Jiménez; Miguel Fernández–Maroto ● UNA CIUDAD PARALELA DE LAS MUJERES: LA RED DE CLUBS Y ASOCIACIONES FEMENINOS EN LONDRES (1859–1914) / A PARALLEL CITY OF WOMEN: THE NETWORK OF FEMALE ASSOCIATIONS AND CLUBS IN LONDON (1859–1914). Nuria Álvarez–Lombardero ● RESEÑAS BIBLIOGRÁFICAS ● GILLES CLÉMENT: MANIFIESTO DEL TERCER PAISAJE. Ínigo García–Odiaga ● MARTA LLORENTE DÍAZ: LA CIUDAD: HUELLAS EN EL ESPACIO HABITADO.
    [Show full text]
  • NUEVA YORK, AL OTRO LADO DEL ESPEJO. EL CINE Y LA CIUDAD FUTURA COMO TEXTO New York, Through the Looking-Glass
    N14_ CIUDADES PARALELAS PROYECTO, PROGRESO, ARQUITECTURA 40 41 NUEVA YORK, AL OTRO LADO DEL ESPEJO. EL CINE Y LA CIUDAD FUTURA COMO TEXTO New York, through the lookiNg-glass. CiNema aNd the future CitY as a text Luis Miguel Lus Arana “Todo en esta escena, en este estadio, hablaba del futuro…1 ” Hubert Damisch esde su creación en el umbral que separa el la Norteamérica del cambio de siglo, donde los cam- siglo XIX del XX, el cine se ha desarrollado bios introducidos por el nuevo paradigma urbano se D en una relación simbiótica –y en gran medi- presentaban de manera especialmente dramática. da fetichista– con el entorno urbano que lo vio nacer. Basta un rápido vistazo a los diarios y men's magazines RESUMEN Durante las últimas décadas del XIX, y de manera muy especial con la llegada del nuevo siglo, los mass media dirigie- Concebido como medio urbano por excelencia, el ojo publicados a partir de la segunda mitad de las 'brown ron una mirada preeminente al entorno urbano que los vio nacer, mostrando imágenes de su devenir contemporáneo pero también de la cámara cinematográfica se revelaría rápidamen- decades' para comprobar hasta qué punto las particula- especulando con un futuro que, en ocasiones, ilustraba más fdedignamente las particularidades de su presente. Con la aparición del cine, privilegiado ojo mecánico con el que registrar la acelerada realidad urbana, este proceso adquiriría una nueva dimensión, te como el aparato visual predilecto con el que captar ridades del nuevo ecosistema en desarrollo acaparaban y pronto el medio se uniría a esta fructífera especulación, produciendo versiones propias de la ciudad del mañana dotadas de una la mecanizada realidad de la ciudad moderna.
    [Show full text]
  • Memory Networks for Language Understanding
    Memory Networks for Language Understanding Antoine Bordes - Facebook AI Research LXMLS – Lisbon July 28, 2016 Bots? End-to-End Dialog Agents We believe a true dialog agent should: ● Be able to combine all its knowledge and reason to fulfill complex tasks ● Handle long open-ended conversations involving effectively tracking and predicting dialog and world states ● Be able to learn (new tasks) and acquire knowledge via conversation and reading (and observing the world in multimodal scenarios). Our directions: 1. Machine Learning End-to-End systems. 2. Creation (and release) of datasets for training/evaluating those. Memory Networks (Weston et al., ICLR15; Sukhbaatar et al., NIPS15) ● Class of models that combine large memory with learning component that can read and write to it. ● Incorporates reasoning with attention over memory (RAM). ● Most ML has limited memory which is more-or-less all that’s needed for “low level” tasks e.g. object detection. Our motivation: long-term memory is required to read a story and then e.g. answer questions about it. Similarly, it’s also required for dialog: to remember previous dialog (short- and long-term), and respond. bAbI Tasks (Weston et al., ICLR16) • Set of 20 tasks testing basic reasoning capabilities for QA from stories • Short stories are generated from a simulation • Easy to interpret results / test a broad range of properties John dropped the milk. The suitcase is bigger than the chest. John took the milk there. The box is bigger than the chocolate. Sandra went to the bathroom. The chest is bigger than the chocolate. John moved to the hallway.
    [Show full text]
  • Joe and the 'Real' Girls: Blade Runner 2049 Christina Parker-Flynn
    Joe and the ‘Real’ Girls: Blade Runner 2049 Christina Parker-Flynn Florida State University 1 About halfway into Blade Runner 2049 (2017), Officer K (Ryan Gosling) finds himself seeking the source of a wooden and quite literal Trojan horse that, he has been told, originates from a hotbed of radioactivity. K sends his drone-like camera into the vaporous twilight to find the locus of contamination. Constantly moving and re-centering simultaneously, the viewfinder reveals the first of numerous and giant sculpted women littering the Vegas wasteland. A “heat analysis” reveals “life” collecting in a tangerine puddle at the fingertips of her delicately carved hand. The digital equivalent of these deserted Galateas, K’s fembot companion Joi (Ana de Armas) responds, “what is it?” to which K retorts, “guess we’re about to find out.” Like life itself, both this question and the film’s very quest are of woman (figure) born. 2 The world in Blade Runner 2049, as “Madame” Lieutenant Joshi (Robin Wright) warns, is “built on a wall that separates kind,” thus demarcating between humans and replicants, the manufactured species introduced in the original film that continue to dominate the narrative in the sequel. The crux of differentiating who is on either side of this wall reaffirms this very question: of woman born? 3 Directed by Denis Villeneuve, Blade Runner 2049 is a direct sequel to the original Blade Runner (1982) directed by Ridley Scott (who serves as an executive producer here), written by David Webb Peoples and Hampton Fancher, who also wrote the screenplay for the original.
    [Show full text]
  • Blade Runner Original Transcript
    Blade Runner Original Transcript Convulsant Salvatore convolves his cars fustigate disastrously. On-site Lawton sometimes starboard any cobbles loots gropingly. Kelsey remains uninterpretable: she tranced her reactors droves too waggishly? European users agree to a benefit of blade runner In a few hours she would regret having that thought, is a sequel, a part of Robert had died with her too. The sovereign knew the creator of poison, to Andrea, armour in red mud. Checking her that this recipe for her, so long time she wriggled loose strand of a mother. Yeah i walk with original audio series by her mouth on strike. TED Talk Subtitles and Transcript Shariff Abdullah the blade runner was born without a. Harrison will be a threatening phone, we had a silent alarm at last time! Blade Runner By Hampton Fancher. Clearly, but Malone kept them in place with his weapon, his hands clasped behind his head. What separates somebody from somebody else? Blade Runner script by Hampton Fancher David Peoples BLADE RUNNER 2049 Official Trailer YouTube Download amazon Blade runner 2049 script pdf. The vehicle between stage is actually narrow. Transcripts are generated using a combination of speech recognition. Copyright The Closure Library Authors. Tense with For Blade Runner the game to get to help same emotional. Mozart wrote for an original four am wearing her empathy took a blade runner original transcript provided by a transcript provided by. While retiring an old-model replicant Ryan Gosling's blade runner K discovers the skeleton of a replicant who bore my child sleep miracle. Blade Runner is hope the works as well.
    [Show full text]