DOCSLIB.ORG

Intro to CBMM

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Intro to CBMM The Center for Brains, Minds and Machines Overview of the Center Second CBMM Summer School, 2015 The Center for Brains, Minds and Machines Vision Accumulated knowledge and technology, now in place, enables a rapid leap in our scientific understanding of intelligence and our ability to replicate intelligence in engineered systems. Mission We aim to create a new field by bringing together computer scientists, cognitive scientists and neuroscientists to work in close collaboration. The new field – the Science and Engineering of Intelligence – is dedicated to developing a computationally centered understanding of human intelligence and to establishing an engineering practice based on that understanding. Why now Machine Learning Computer Science Neuroscience Computational Cognitive Science Neuroscience Science+ Technology of Intelligence3 Centerness: collaborations across different disciplines and labs MIT Harvard Boyden, Desimone ,Kaelbling , Kanwisher, Blum, Kreiman, Mahadevan, Katz, Poggio, Sassanfar, Saxe, Nakayama, Sompolinsky, Schulz, Tenenbaum, Ullman, Wilson, Spelke, Valiant Rosasco, Winston Rockefeller Allen Institute UCLA Stanford Cornell Freiwald Koch Yuille Goodman Hirsh Hunter Wellesley Puerto Rico Howard Epstein,Sakas, Hildreth, Conway, Bykhovaskaia, Ordonez, Manaye, Chouikha, Chodorow Wiest Arce Nazario Rwebargira CBMM Organizational Chart Dean of the School of Science, MIT Robert Desimone, Director, McGovern Institute for Brain Research at MIT Education Evaluation Director External Advisory Committee UIUC: Lizanne DeStefano MIT: Tomaso Poggio Center Director Associate Director Associate Director Managing Director MIT: Kathleen HU: L Mahadevan MIT: Matt Wilson HU: Kenneth Blum Sullivan HU: Liz Spelke Education Coordinator KT Coordinator Research Coordinator Diversity Coordinator WC: Ellen Hildreth MIT: Boris Katz MIT: Patrick Winston MIT: Mandana Sassanfar CU: Haym Hirsh HU: L Mahadevan T3 Visual T1 Developing T2 Circuits for T4 Social T5 Theory for Intelligence Intelligence Intelligence Intelligence Intelligence MIT: Shimon MIT: Josh HU: Gabriel MIT: Nancy MIT: Tomaso Ullman Tenenbaum Kreiman Kanwisher Poggio MIT: Boris Katz The Center for Brains, Minds and Machines A little bit of history and background 6 Why now: recent progress in AI Why now: very recent progress in AI 10 11 Why now: very recent progress in AI 14 15 Thus we see great advances in AI based on machine learning research of 20 years ago … but we are still very far from understanding human intelligence and the brain 16 What is this? What is Hueihan doing? What does Hueihan think about Joel’s thoughts about her? Intelligence and Turing++ Questions • Intelligence —> Human Intelligence • (Human) Intelligence: one word, many problems • A CBMM mission: define and “answer” these Turing++ Questions 18 Turing++ Questions functi onal theor y CB M M th eo ry The challenge is to develop computational models that answer questions about images and videos such as what is there / who is there / what is the person doing and eventually more difficult questions such as who is doing what to whom? • what happens next? at the computational, psychophysical and neural levels. The who question: face recognition from experiments to theory (Workshop, Sept 4-5, 2015) Model ML AL AM Neural Circuits of Thrust 5 Intelligence Social Visual Intelligence Intelligence Thrust 1 More about CBMM Second CBMM Summer School, 2015 Industrial partners IIT A*star Hebrew U. MPI Metta, Rosasco, Shashua Buelthoff Sandini Tan Genoa U. Weizmann City U. HK NCBS Smale Verri Ullman Raghavan Google IBM Microsoft Siemens Schlumberger GE Norvig Lemnios Blake Boston DeepMind Orcam Rethink Robotics MobilEye Dynamics Shashua Shashua Hassabis Raibert Brooks Adding value in Reserach Thrust 1: Thrust 2: Development of Circuits for Intelligence Intelligence Josh Tenenbaum Gabriel Kreiman Thrust 5: Theory of Intelligence Tomaso Poggio Thrust 3: Thrust 4: Visual Intelligence Social Intelligence Shimon Ullman Nancy Kanwisher a. Thrust 1 The goal of Thrust 1 is to understand the roots of human intelligence o Infants’ early physical representations: Forces and masses o Learning physics from dynamic scenes o Unifying frameworks for intuitive physics beyond rigid, solid objects o Efficiency and experience in infants’ goal representations o Computing Efficiency in Infancy o Efficiency and effort: Bridging intuitive psychology and intuitive physics o Social cognition and the naïve utility calculus o Do children expect others to learn rationally from evidence? o Inference of mental states from emotional expressions o How do children evaluate potential social partners? o Early reasoning about affiliation and kinship o Children’s perception of causal relations across intuitive physics and psychology o Knowing what to look for, in object perception and exploratory learning o Building more sophisticated models of children’s learning in goal-directed search o Lookit: An online developmental lab o Probabilistic programs for scene understanding o Hierarchical motion perception b. Thrust 2 Thrust 2 seeks to discover the neural circuits that answer Turing++ Questions such as “What and who are there? What are they doing?” To achieve this goal, scientists in Thrust 2 record electrical signals from the brains of humans and experimental animals while they are identifying objects and interpreting scenes, and then build computational models of the circuit function. • Towards a biologically inspired human-level language processing system • Decoding human action and interaction in the human brain • Top-down processes in extracting meaning from images • Task-dependence in visual processing • Neural mechanisms of face detection • Neural circuits underlying navigation • Computational architectures mediating object recognition under occlusion • Biological and computational code underlying pattern completion • Light field Neural Imaging • Expansion Microscopy • Scalable Neural Recording Probes c. Thrust 3 Thrust 3 aims to determine the elements involved in fully understanding visual scenes. Our computational models should be able to extract from the scene information about actions, agents, goals, object configurations, social interactions, and more. • Answering questions about videos • Disambiguating language and video • Learning Vision and Language with weak supervision • Learning attention mechanisms and segmentation • Visual processing with minimal recognizable configurations • Human language learning • Understanding intentions and goals in human interactions from gaze direction • Towards a biologically inspired human-level language processing system • Dataset development d. Thrust 4 The goal of Thrust 4 is to understand the capabilities and brain-bases of various aspects of social perception. • Functional organization of social perception and cognition in the superior temporal sulcus • Face Looking Behavior and Image Statistics for Faces in the Real World • Fast reading of Action Goals • Decoding human action and interaction in the human brain • Understanding intentions and goals in human social interactions from gaze direction e. Thrust 5 The Theoretical Thrust aims to provide understanding, guide computer implementations, and inform and interpret experiments for the other thrusts. • Biologically Plausible Implementations of i-theory for face and object recognition • Invariance and Selectivity in Visual Cortex • Invariant Representation Learning for Speech Recognition • Invariant representations for action recognition in the human visual system • Learning and Reasoning in Symbolic Domains • Probabilistic Learning and Inference in a Neuroidal Model of Computation • The computational role in object recognition of eccentricity dependent resolution in retina • Computational architectures mediating object recognition with occlusion (with Thrust 2) • The development of geometrical intuitions • Visual routines • Compositional Models of Humans and Animals • Learning Vision and Language with weak supervision 28 A summary To understand (human) intelligence, we must: • Understand what we compute • How what we compute develops • How amplified by social interaction • How implemented in neural tissues.
Load more

Recommended publications
  • Tomaso A. Poggio
    BK-SFN-NEUROSCIENCE-131211-09_Poggio.indd 362 16/04/14 5:25 PM Tomaso A. Poggio BORN: Genova, Italy September 11, 1947 EDUCATION: University of Genoa, PhD in Physics, Summa cum laude (1971) APPOINTMENTS: Wissenschaftlicher Assistant, Max Planck Institut für Biologische Kybernetik, Tubingen, Germany (1978) Associate Professor (with tenure), Department of Psychology and Artificial Intelligence Laboratory, Massachusetts Institute of Technology (1981) Uncas and Helen Whitaker Chair, Department of Brain & Cognitive Sciences, Massachusetts Institute of Technology (1988) Eugene McDermott Professor, Department of Brain and Cognitive Sciences, Computer Science and Artificial Intelligence Laboratory and McGovern Institute for Brain Research, Massachusetts Institute of Technology (2002) HONORS AND AWARDS (SELECTED): Otto-Hahn-Medaille of the Max Planck Society (1979) Member, Neurosciences Research Program (1979) Columbus Prize of the Istituto Internazionale delle Comunicazioni Genoa, Italy (1982) Corporate Fellow, Thinking Machines Corporation (1984) Founding Fellow, American Association of Artificial Intelligence (1990) Fellow, American Academy of Arts and Sciences (1997) Foreign Member, Istituto Lombardo dell’Accademia di Scienze e Lettere (1998) Laurea Honoris Causa in Ingegneria Informatica, Bicentenario dell’Invezione della Pila, Pavia, Italia, March (2000) Gabor Award, International Neural Network Society (2003) Okawa Prize (2009) Fellow, American Association for the Advancement of Science (2009) Tomaso Poggio began his career in collaboration
    [Show full text]
  • How Visual Cortex Works and Machine Vision Tools for the Visually Impaired
    ! How visual cortex works and machine vision tools for the visually impaired ! tomaso poggio CBMM, McGovern Institute, BCS, CSAIL MIT The Center for Brains, Minds and Machines Vision for CBMM • The problem of intelligence is one of the great problems in science. • Work so far has led to many systems with impressive but narrow intelligence • Now it is time to develop a basic science understanding of human intelligence so that we can take intelligent applications to another level. MIT Harvard Boyden, Desimone ,Kaelbling , Kanwisher, Blum, Kreiman, Mahadevan, Katz, Poggio, Sassanfar, Saxe, Nakayama, Sompolinsky, Schulz, Tenenbaum, Ullman, Wilson, Spelke, Valiant Rosasco, Winston Cornell Hirsh Allen Ins1tute Rockfeller UCLA Stanford Koch Freiwald Yuille Goodman Hunter Wellesley Puerto Rico Howard Epstein,... Hildreth, Conway... Bykhovaskaia, Vega... Manaye,... City U. HK Hebrew U. IIT MPI Smale Shashua MeNa, Rosasco, Buelthoff Sandini NCBS Genoa U. Weizmann Raghavan Verri Ullman Google IBM MicrosoH Orcam MoBilEye Norvig Ferrucci Blake Shashua Shashua Boston Rethink Willow DeepMind Dynamics RoBo1cs Garage Hassabis Raibert Brooks Cousins RaPonal for a Center Convergence of progress: a key opportunity Machine Learning & Computer Science Neuroscience & Computational Cognitive Science Neuroscience Change comes most of all from the unvisited no-man’s-land between the disciplines (Norbert Wiener (1894-1964)) Science + Technology of Intelligence The core CBMM challenge:describe in a scene objects, people, actions, social interactions What is this?
    [Show full text]
  • Old HMAX and Face Recognition
    Old HMAX and face recognition Tomaso Poggio 1. Problem of visual recognition, visual cortex 2. Historical background 3. Neurons and areas in the visual system 4. Feedforward hierarchical models 5. Beyond hierarchical models WARNING: using a class of models to summarize/interpret experimental results ! • Models are cartoons of reality, eg Bohr’s model of the hydrogen atom ! • All models are “wrong” ! • Some models can be useful summaries of data and some can be a good starting point for a real theory 1. Problem of visual recognition, visual cortex 2. Historical background 3. Neurons and areas in the visual system 4. Feedforward hierarchical models 5. Beyond hierarchical models Learning and Recogni-on in Visual Cortex: what is where Unconstrained visual recognition was a difficult problem (e.g., “is there an animal in the image?”) Vision: what is where ! ! ! Vision A Computational Investigation into the Human Representation and Processing of Visual Information David Marr Foreword by Shimon Ullman !Afterword by Tomaso Poggio David Marr's posthumously published Vision (1982) influenced a generation of brain and cognitive scientists, inspiring many to enter the field. In Vision, Marr describes a general framework for understanding visual perception and touches on broader questions about how the brain and its functions can be studied and understood. Researchers from a range of brain and cognitive sciences have long valued Marr's creativity, intellectual power, and ability to integrate insights and data from neuroscience, psychology, and computation. This MIT Press edition makes Marr's influential work available to a new generation !of students and scientists. In Marr's framework, the process of vision constructs a set of representations, starting from a description of the input image and culminating with a description of three-dimensional objects in the surrounding environment.
    [Show full text]
  • Learning from Brains How to Regularize Machines
    Learning From Brains How to Regularize Machines Zhe Li1-2,*, Wieland Brendel4-5, Edgar Y. Walker1-2,7, Erick Cobos1-2, Taliah Muhammad1-2, Jacob Reimer1-2, Matthias Bethge4-6, Fabian H. Sinz2,5,7, Xaq Pitkow1-3, Andreas S. Tolias1-3 1 Department of Neuroscience, Baylor College of Medicine 2 Center for Neuroscience and Artificial Intelligence, Baylor College of Medicine 3 Department of Electrical and Computer Engineering, Rice University 4 Centre for Integrative Neuroscience, University of Tübingen 5 Bernstein Center for Computational Neuroscience, University of Tübingen 6 Institute for Theoretical Physics, University of Tübingen 7 Institute Bioinformatics and Medical Informatics, University of Tübingen *[email protected] Abstract Despite impressive performance on numerous visual tasks, Convolutional Neural Networks (CNNs) — unlike brains — are often highly sensitive to small pertur- bations of their input, e.g. adversarial noise leading to erroneous decisions. We propose to regularize CNNs using large-scale neuroscience data to learn more robust neural features in terms of representational similarity. We presented natural images to mice and measured the responses of thousands of neurons from cortical visual areas. Next, we denoised the notoriously variable neural activity using strong predictive models trained on this large corpus of responses from the mouse visual system, and calculated the representational similarity for millions of pairs of images from the model’s predictions. We then used the neural representation similarity to regularize CNNs trained on image classification by penalizing intermediate representations that deviated from neural ones. This preserved performance of baseline models when classifying images under standard benchmarks, while main- taining substantially higher performance compared to baseline or control models when classifying noisy images.
    [Show full text]
  • A Framework for Searching Forgeneral Artificial Intelligence
    M.ROSA,J.FEYEREISL&THEGOODAITEAM AFRAMEWORK FORSEARCHINGFOR GENERALARTIFICIALINTELLIGENCE arXiv:1611.00685v1 [cs.AI] 2 Nov 2016 V E R S I O N 1 Versions 1.0 current version: initial release for the general public 2.0 planned future version: additional illustrations demonstrating framework principles 3.0 planned future version: additional mathematical formalizations and proofs Correspondence: {marek.rosa,jan.feyereisl}@goodai.com Copyright © 2016 GoodAI www.goodai.com The LATEX template used in this document is licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0. Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “as is” basis, without warranties or conditions of any kind, either express or implied. See the License for the specific language governing permissions and limitations under the License. First printing, November 2016 Abstract There is a significant lack of unified approaches to building generally intelligent machines. The majority of current artificial intelligence re- search operates within a very narrow field of focus, frequently without considering the importance of the ’big picture’. In this document, we seek to describe and unify principles that guide the basis of our de- velopment of general artificial intelligence. These principles revolve around the idea that intelligence is a tool for searching for general so- lutions to problems. We define intelligence as the ability to acquire skills that narrow this search, diversify it and help steer it to more promising areas.
    [Show full text]
  • Cognitive Architectures for Robot Agents Current Capabilities, Future Enhancements and Prospects for Collaborative Development CONTENT WELCOME
    Cognitive Architectures for Robot Agents Current Capabilities, Future Enhancements and Prospects for Collaborative Development CONTENT WELCOME WELCOME 3 From March 22 to March 28, 2020, researchers from PANEL DISCUSSIONS 4 Europe, the United States and various countries around the globe met in a virtual venue to discuss Learning and Adaptation in the Minimalist Cognitive Architecture 4 the current capabilities and future prospects of cog- Understanding Human Cognition through Modelling the Human Brain 7 nitive architectures. Developmental Cognitive Robotics as a Gateway to a More Natural HRI 10 As we are building increasingly powerful AI systems PRESENTATION SUMMARIES 14 and robot agents, we also feel more pressure to im- Minimalist Cognitive Architectures 14 prove the orchestration of representations and com- Collaborating on Architectures: Challenges and Perspectives 15 putation processes required for achieving competent agency. Probably one of the biggest lessons we have Cognitive Architectures for Assistive Robot Agents 16 learned over the last 30 to 50 years is the role that ArmarX – A Robot Cognitive Architecture 17 actions play in intelligent agency. In the beginning, Affective Architecture: Pain, Empathy, and Ethics 18 many considered actions to be atomic entities that Cognitive Robotics and Control 19 can be modeled in a fairly simple way. We tried to Circuits for Intelligence 20 realize competent agency through reasoning about The LIDA Cognitive Architecture – An Introduction with Robotics Applications 21 these action models. Clarion: A comprehensive, Integrative Cognitive Architecture 22 I think this view has shifted significantly. Many of us The workshop showcased the huge potential and The DIARC Architecture for Autonomous Interactive Robots 23 feel that the key to competent agency can be found large synergies that we can unlock if we work to- The Soar Cognitive Architecture: Current and Future Capabilities 24 inside the actions.
    [Show full text]
  • TOMASO A. POGGIO Department of Brain & Cognitive
    TOMASO A. POGGIO Department of Brain & Cognitive Sciences McGovern Institute for Brain Research Computer Science & Artificial Intelligence Lab Massachusetts Institute of Technology 43 Vassar Street Cambridge, MA 02142 URL: http://cbcl.mit.edu/people/poggio/poggio-cv-web.htm Date of Birth: September 11, 1947 Citizenship: U.S. EDUCATION Ph.D. in Theoretical Physics, University of Genoa (1970), Summa cum Laude. Thesis: “On Holographic Models of Memory.” ACADEMIC POSITIONS Professor, Department of Brain & Cognitive Sciences, Computer Science & Artificial Intelligence Laboratory and McGovern Institute for Brain Research, Massachusetts Institute of Technology, 1984-present. Associate Professor, Department of Psychology and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, 1981-1984. Wissenschaftlicher Assistant, Max Planck Institut für Biologische Kybernetik, Tubingen, Germany, 1971-1981. HONORS AND SERVICE Honorary Chair: European Conference on Computer Vision (ECCV), Florence, Italy, October 7-13, 2012 Valedictory Talk: “The Future of the Science and Engineering of Intelligence,” IEEE CIFER 2012, New York City, March 30, 2012 Honorary Chair: IEEE Computational Intelligence for Financial Engineering and Economics (CIFEr) 2012, March 29-30, 2012, New York City Member of ISICT Committee of Experts, 2010 Member of the Duke University External Review Committee (Dept. CS), 2010 General Chair, 2010 International Conference on Brain Informatics (BI 2010), August 28-30, 2010, Toronto, Canada American Association for the Advancement of Science (AAAS) Fellow (2009) Okawa Prize, 2009 Keynote address at V incontro annuale ISICT, Genoa, Italy, October 2009 Honorary Member of EEE Symposium on Computational Intelligence for Financial Engineering (CIFEr 2009) Member of the Scientific Board of ISI Turin, 2006 – present Co-organizer of Workshop on Learning Theory, FOCM ’05, Santander, Spain, 2005.
    [Show full text]
  • Curriculum Vitae | Page 1 of 25
    Thomas Serre, PhD Curriculum Vitae | Page 1 of 25 Thomas Serre, PhD Brown University | Carney Institute for Brain Science Associate Professor, Cognitive, Linguistic & Psychological Sciences Department http://serre-lab.clps.brown.edu Education 2001 – 2006 Ph.D. in Neuroscience Mass. Institute of Technology (Cambridge, MA), Brain and Cognitive Sciences Department Thesis title: Learning a dictionary of shape-components in visual cortex: Comparison with ​ neurons, humans and machines. Advisor: Prof. Tomaso Poggio ​ ​ ​ ​ 1999 – 2000 MSc in Statistics and Probability Theory Université de Rennes (Rennes, France) 1997 – 2000 MSc in EECS Ecole Nationale Supérieure des Télécommunications de Bretagne (Brest, France) Major in image processing 1995 – 1997 BSc in Math and Physics (Classes préparatoires aux Grandes Ecoles) Lycée Pasteur (Neuilly, France) Professional appointments 2020– Associate Director Center for Computational Brain Science (Brown University) 2019– International Chair in AI ANR-3IA Artificial and Natural Intelligence Toulouse Institute (Toulouse, France) 2018 – Faculty Director Brown University, Center for Computation and Visualization (Brown University) 2017 – Associate Professor (tenured) Brown University, Department of Cognitive, Psychological and Linguistic Sciences 2013 – 2017 Manning Assistant Professor Brown University, Department of Cognitive, Psychological and Linguistic Sciences 2011 – Associate Director Carney Behavioral Phenotyping Core Facility (Brown University) Last updated: 03/15/2020 Thomas Serre, PhD Curriculum Vitae | Page 2 of 25 2010 – 2013 Assistant Professor Brown University, Department of Cognitive, Psychological and Linguistic Sciences 2006 – 2009 Postdoctoral Associate Mass. Institute of Technology (Cambridge, MA), McGovern Institute for Brain Research Completed Publications A list of publications and citations can be found on my Google Scholar profile at https://scholar.google.com/citations?user=kZlPW4wAAAAJ&hl=en.
    [Show full text]
  • Minimal Images in Deep Neural Networks: Fragile Object Recognition in Natural Images
    Published as a conference paper at ICLR 2019 MINIMAL IMAGES IN DEEP NEURAL NETWORKS: FRAGILE OBJECT RECOGNITION IN NATURAL IMAGES Sanjana Srivastava1;3, Guy Ben-Yosef1;2,∗ Xavier Boix1;3;4∗ 1Center for Brains, Minds, and Machines (CBMM) 2Computer Science and Artificial Intelligence Laboratory, Massachusetts Institute of Technology, USA 3Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, USA 4Children’s Hospital, Harvard Medical School, USA {sanjanas, gby, xboix}@mit.edu ABSTRACT The human ability to recognize objects is impaired when the object is not shown in full. "Minimal images" are the smallest regions of an image that remain recog- nizable for humans. Ullman et al. (2016) show that a slight modification of the location and size of the visible region of the minimal image produces a sharp drop in human recognition accuracy. In this paper, we demonstrate that such drops in accuracy due to changes of the visible region are a common phenomenon between humans and existing state-of-the-art deep neural networks (DNNs), and are much more prominent in DNNs. We found many cases where DNNs classified one region correctly and the other incorrectly, though they only differed by one row or column of pixels, and were often bigger than the average human minimal image size. We show that this phenomenon is independent from previous works that have reported lack of invariance to minor modifications in object location in DNNs. Our results thus reveal a new failure mode of DNNs that also affects humans to a much lesser degree. They expose how fragile DNN recognition ability is for natural images even without adversarial patterns being introduced.
    [Show full text]
  • Levels of Analysis for Machine Learning
    Published as a workshop paper at “Bridging AI and Cognitive Science” (ICLR 2020) LEVELS OF ANALYSIS FOR MACHINE LEARNING Jessica B. Hamrick Shakir Mohamed DeepMind DeepMind [email protected] [email protected] ABSTRACT Machine learning is currently involved in some of the most vigorous debates it has ever seen. Such debates often seem to go around in circles, reaching no conclusion or resolution. This is perhaps unsurprisinggiven that researchers in machine learn- ing come to these discussions with very different frames of reference, making it challenging for them to align perspectives and find common ground. As a remedy for this dilemma, we advocate for the adoption of a common conceptual frame- work which can be used to understand, analyze, and discuss research. We present one such framework which is popular in cognitive science and neuroscience and which we believe has great utility in machine learning as well: Marr’s levels of analysis. Througha series of case studies, we demonstrate howthe levels facilitate an understanding and dissection of several methods from machine learning. By adopting the levels of analysis in one’s own work, we argue that researchers can be better equipped to engage in the debates necessary to drive forward progress in our field. 1 CONCEPTUAL FRAMEWORKS The last few months and years have seen researchers embroiled in heated debates touching on funda- mental questions in machine learning: what, exactly, is “deep learning”? Can it ever be considered “symbolic”? Is it sufficient to achieve artificial general intelligence? Can it be trusted in high-stakes, real-world applications? These are important questions, yet the discussions surrounding them seem unable to move forward productively.
    [Show full text]
  • The Science of Deep Learning COLLOQUIUM INTRODUCTION Richard Baraniuka, David Donohob,1, and Matan Gavishc
    COLLOQUIUM INTRODUCTION The science of deep learning COLLOQUIUM INTRODUCTION Richard Baraniuka, David Donohob,1, and Matan Gavishc artificial intelligence | machine learning | deep learning | neural networks Scientists today have completely different ideas of what people posted trillions of images and documents on machines can learn to do than we had only 10 y ago. social media, as the phrase “Big Data” entered media In image processing, speech and video processing, awareness. Image processing and natural language machine vision, natural language processing, and clas- processing were forever changed by this new data sic two-player games, in particular, the state-of-the-art resource as they tapped the new image and text re- has been rapidly pushed forward over the last decade, sources using the revolutionary increases in comput- as a series of machine-learning performance records ing power and newly globalized human talent pool. were achieved for publicly organized challenge prob- The field of image processing felt the impact of the lems. In many of these challenges, the records now new data first, as the ImageNet dataset scraped from meet or exceed human performance level. the web by Fei-Fei Li and her collaborators powered a A contest in 2010 proved that the Go-playing com- series of annual ImageNet Large-Scale Visual Recog- puter software of the day could not beat a strong hu- nition Challenge (ILSVRC) prediction challenge com- man Go player. Today, in 2020, no one believes that petitions. These competitions gave a platform for the human Go players—including human world champion emergence and successive refinement of today’s Lee Sedol—can beat AlphaGo, a system constructed deep-learning paradigm in machine learning.
    [Show full text]
  • The Course of the Mind. Steven Ashley
    The Course of the Mind By Steven Ashley One of the greatest mysteries of nature is the human mind. Understanding how the brain produces intelligent behavior, and how that might be replicated by machines, are among the greatest problems facing science and technology. Even on a solely personal level, consider: What’s more interesting to you than you and how you work? Nothing, except perhaps the prospect of a “robot-you” to help you live and work better. Yet despite decades of study, the three-pound bundle of moist gray matter inside your skull has remained a black box, for the most part a padlocked one. Although the early computer scientists were optimistic in the mid-1950s when “artificial intelligence” began growing into an academic field, succeeding years taught them that emulating the operations of 100 billion highly evolved, networking neurons is much harder than it first seemed. Harder than winning championship chess matches, it turned out, or teaching machines to recognize voices and translate texts by analyzing huge numbers of labeled training examples, as do Apple’s Siri and IBM’s Watson. “These recent achievements only underscore the limitations of computer science and artificial intelligence,” says Tomaso Poggio, professor of brain sciences and human behavior at MIT and co-founder of the Brains, Minds and Machines course at MBL, which launched in the summer of 2014. “We do not yet understand how the brain gives rise to intelligence, nor do we know how to build machines that are as broadly intelligent as we are,” he says. Poggio, who heads up the Center for Brains, Minds and Machines (CBMM) at MIT, thinks it’s high time to renew the quest to achieve the field’s grand goal to explain the brain.
    [Show full text]