I focus on the basics and the motivations.
I draw pictures to make things easy and motivate them.
The courses are for a broad range-from any topics in the sixth grade to certain undergraduate topics in mathematics, e.g. calculus of single or multiple variables, geometry, group theory, linear algebra etc.
I'm a PhD in Mathematics, and currently work in artificial intelligence. Feel free to ask for my credentials, or see my profiles on LinkedIn or ResearchGate. I also have undergraduate in both mathematics and statistics, and I've published in both mathematics and artificial inteligence-my papers are on ResearchGate.
The payment can be paid just in the beginning of the lesson, either in cash or by bank transfer, in the second case, I'll give you my bank detail.
One more thing: I'll give the courses in English only even though I can speak French to a certain extent. The reason is: I don't know the technical terms for mathematics in French.
Connect Things, France July 2018 — September 2018
Machine learning researcher
Ecole Polytechnique, France November 2016 — December 2017
Research scientist in mathematical computer vision
University of California at Los Angeles, USA
February 2016 — November 2016
Research scientist in medical image processing
INRIA, Paris, France October 2014 — October 2015
Research scientist in medical image processing
Doctor of Philosophy (Ph.D.) in Mathematics
September 2007 — September 2013
Rutgers University, New Brunswick, NJ, USA
Research in differential geometry and low dimensional topology with internationally published papers.
Master of Science in Applied Mathematics
June 2005 — June 2007
Tata Institute of Fundamental Research, Bangalore, India
Courses in linear algebra, statistics, advanced probability, differential geometry, numerical analysis, differential equations, algebraic topology.
1. Classification of fetal alcohol syndrome patients (coding in
MATLAB): Performed linear discriminant analysis (LDA) on 175 precentral and central sulcal curves, with a 120-dimensional feature space consisting of Fourier coefficients of the curves. Estimated the performance of this classifier by a 10-fold cross validation to obtain classification accuracy of 76%.
⦁ Image classification by their projections and shape difference matrices (coding in MATLAB and Python): We apply support vector machines (SVM) and linear discriminant analysis (LDA) to the HOG-features of 2D projections from 4 different camera angles of 3D images of human and non-human data, and initially obtained 98% test accuracy. Since number of HOG features (1764) here exceeded the number of samples (1524), we perform regularized LDA to select best 588 features, and still obtain 95% test accuracy. The end goal of the project is to see how adding information from shape difference matrices effect the accuracy.
⦁ Development of a new mixed effect longitudinal model for Alzheimer's disease
modeling (coding in MATLAB): Developed a new mixed effect longitudinal model on shape spaces relevant to medical imaging and computer vision, using the geometric concept of parallel transport, that compares time series trajectories with different onsets. I developed and implemented an algorithm to estimate parallel transport with error rate O(1/number of steps), showed the error rate is the best achievable, and also has a smaller error compared to Schild's ladder algorithm, previously used to compute parallel transport.
⦁ Clustering of retail product sizes (coding in Python and OpenCV): Using the heights and widths obtained from the bounding box co-ordinates of the products from the detection models, I used the feature vector (height, width, square root of size) of the image bounding box co-ordinates to cluster product sizes themselves, with bigger sizes getting higher cluster numbers, using both KMeans and MeanShift algorithms. With no information on camera distances, achieved a clustering accuracy of around 75%. The next step is to train a suitable classification algorithm based on this clustering, and test it for predicting sizes.
⦁ Determining popular places from GPS and other geolocation related data (coding in Python): I constructed a suitable cost function which is an indicator of how popular a co-ordinate of earth's surface is, in the sense that it measures how many trajectories pass at or near that co-ordinate or through the corresponding geohash. By minimizing this cost function, which is further a function of distance to trajectories, I was able to estimate the popularity of a place.
MACHINE LEARNING AND MATHEMATICS PAPERS:
⦁ Construction of a closed hyperbolic surface of arbitrarily small eigenvalue of prescribed serial number. Contemporary Mathematics, 2011.
⦁ (With Jun Hu) Boundary differentiability of Douady-Earle extensions of diffeomorphisms of S^n. Pure and Applied Mathematics Quarterly, 2013.
⦁ (With Jun Hu) Douady-Earle extensions of Holder continuous and Lipschitz continuous circle homeomorphisms. Preprint.
⦁ (With Stanley Durrleman et. al.) A Fanning Scheme for Parallel Transport along Geodesics on Riemannian manifolds. Siam Journal of Numerical Analysis, 2018.
⦁ (With Praneeth Vepakomma) Optimal bandwidth estimation for dimensional reduction to detect topological circularity in high dimensional data. Submitted.
⦁ (With Shantanu Joshi et. al.) Linear and Quadratic Discriminant Analysis on the tangent space of curve-valued data. Differential Geometry in Computer Vision and Machine Learning, 2017. Best paper award.
MACHINE LEARNING SKILLS:
SVM, decision trees, discriminant analysis, linear and logistic regressions, PCA, regularization, cross validation, Markov and hidden Markov models, clustering. Theoretical familiarity with feedforward neural networks (fully connected, convolutional, backpropagation), stochastic gradient descent.
Matlab (~2 years), Python: NumPy, sk-learn, Pandas, OpenCV (4 months), Some familiarity with Linux, LaTeX.
English (bilingual), French (intermediate), Bengali (native), Hindi (intermediate).
Available upon request, but you can also see the ones on LinkedIn.
des cours de maths à proximité ? Voici une sélection d'annonces de professeurs pouvant vous accompagner. Superprof peut aussi vous proposer des cours d'algèbre pour vous aider. Apprendre n'est plus un soucis, des cours de géométrie accessible(s) pour tous ! Suivre des cours de trigonométrie n'a jamais été aussi facile : vous allez aimer développer de nouvelles connaissances.
|Chez lui||Chez vous||Par webcam|