Statistical Learning Theory and ApplicationsLearning: Brains and MachinesLearning: much more than memoryLearning: math, engineering, neuroscienceLearning: math, engineering, neuroscience (now) Math Camp? Look at old Mathcamps on Web site: we will decide on Monday9.520 Statistical Learning Theory and Applications (2007) 9.520 Statistical Learning Theory and Applications Project suggestionsOverview of overviewReason to learn some learning theoryNotesSlide Number 13 Learning from examples: goal is not to memorize but to generalize, eg predict.Binary classification caseSlide Number 16Slide Number 17 A key requirement for learning: generalization Slide Number 19Slide Number 20Slide Number 21 Well-posedness and generalization: are they related?Slide Number 23Slide Number 24Regularization in RKHS: a simple algorithm which generalizes well (it is uniformly stable) and is computationally tractableSlide Number 26Winning against the curse of dimensionality: new research directions in learningSlide Number 28Slide Number 29Overview Overview of overviewLearning from Examples: engineering applicationsObject recognition for computer vision: (personal) historical perspectiveLearning from Examples: engineering applicationsLearning Object Detection: Finding Frontal Faces ...Learning Face DetectionFace detection:…Learning from Examples: engineering applicationsSlide Number 39We did well with shallow learning architectures (SVMs): ~10 year old CBCL computer vision work: SVM-based pedestrian detection system in Mercedes test car… now becoming a product (MobilEye, Israeli company)Slide Number 41Remark: training set defines task People classification/detectionSlide Number 43Slide Number 44Learning from Examples: engineering applications The street scene projectSlide Number 47Slide Number 48Slide Number 49Learning from Examples: engineering applicationsThe problem: action recognitionA new model of the dorsal stream (motion) following the ventral stream model Using a large dictionary of MT-like units for action recognition works well!Learning from Examples: engineering applications Can we “read-out” from visual cortex what the monkey sees?Slide Number 56Slide Number 57Slide Number 58 Training a classifier on neuronal activity.Slide Number 60Learning from Examples: engineering applicationsSlide Number 62Slide Number 63Learning from Examples: engineering applicationsImage AnalysisImage SynthesisReconstructed 3D Face Models from 1 imageReconstructed 3D Face Models from 1 imageSlide Number 69Extending the same basic learning techniques (in 2D): Trainable Videorealistic Face Animation (voice is real, video is synthetic)Trainable Videorealistic Face AnimationSlide Number 72A Turing test: what is real and what is synthetic?Overview of overviewSlide Number 75 A hierarchical model of the ventral stream, which is also a (unsupervised + supervised) learning algorithm… …predicts and is consistent with neural data… …mimics human recognition performance in rapid categorization (and does well as on computer vision benchmarks)Slide Number 79Slide Number 80It is just possible that the brain ….Overview of overviewStatistical Learning Theory and ApplicationsLorenzo Rosasco + Jake Bouvrie +Ryan Rifkin + Charlie Frogner + Tomaso PoggioMcGovern Institute for Brain ResearchCenter for Biological and Computational LearningDepartment of Brain & Cognitive SciencesMassachusetts Institute of TechnologyCambridge, MA 02139 USA9.250 in 2009Learning: Brains and MachinesLearning is the gateway to understanding the brain and to making intelligent machines. Problem of learning: a focus for o modern matho computer algorithmso neuroscienceLearning: much more than memory• Role of learning (theory and applications in many different domains) has grown substantially in CS: learning+statistics is becoming a lingua franca in CS• Plasticity and learning increasingly have a central stage in the neurosciences• Until now math and engineering of learning has developed independently of neuroscience…but it may begin to change as we will see in the class.Learning theory+ algorithmsComputationalNeuroscience: models+experimentsENGINEERING APPLICATIONS• Bioinformatics• Computer vision• Computer graphics, speech synthesis, creating a virtual actorHow visual cortex works – and how it may suggest better computer vision systemsLearning: math, engineering, neuroscience211min ( , ( ))iiKfHiVy fx fμ∈=⎡⎤+⎢⎥⎣⎦∑llTheorems on foundations of learning:Predictive algorithmsLearning theory+ algorithmsComputationalNeuroscience: models+experimentsENGINEERING APPLICATIONSHow visual cortex works:Deep Learning in Cortex Learning: math, engineering, neuroscience (now)211min ( , ( ))iiKfHiVy fx fμ∈=⎡⎤+⎢⎥⎣⎦∑llTheorems on foundations of learningPredictive algorithms(Regularization networks ~ SMS…)• Bioinformatics• Computer vision • Computer graphics, speech synthesis• Speech recognitionMath Camp? Look at old Mathcamps on Web site: we will decide on MondayFunctional Analysis:Linear and Euclidean spacesscalar product, orthogonalityorthonormal bases, norms and semi-norms,Cauchy sequence and complete spacesHilbert spaces, function spaces and linear functional, Riesz representation theorem, convex functions, functional calculus.Probability Theory:Random Variables (and related concepts), Law of Large Numbers, Probabilistic Convergence, Concentration Inequalities.Linear AlgebraBasic notion and definitions: matrix and vectors norms, positive, symmetric, invertible matrices, linear systems, condition number.& Multivariate Calculus:Extremal problems, differential, gradient.9.520 Statistical Learning Theory and Applications (2007) Class 26: Project presentations (past examples)10:30 - Simon Laflamme “Online Learning Algorithm for Structural Control using Magnetorheological Actuators”- Emily Shen “Time series prediction”- Zak Stone “Facebook project”- Jeff Miller “Clustering features in the standard model of cortex”- Manuel Rivas "Learning Age from Gene Expression Data“- Demba Ba “Sparse Approximation of the Spectrogram via Matching Pursuits: Applications to Speech Analysis”- Nikon Rasumov "Data mining in controlled environment and real data"9.520 Statistical Learning Theory and Applications Class 26: Project presentations (past examples)2:35-2:50 "Learning card playing strategies with SVMs", David Craft and Timothy Chan2:50-3:00 "Artificial