Open Hub
Open Hub

 Settings |  Report Duplicate

0
I Use This!
Very Low Activity

Commits : Listings

Analyzed about 9 hours ago. based on code collected 1 day ago.
Nov 05, 2025 — Nov 05, 2026
Commit Message Contributor Files Modified Lines Added Lines Removed Code Location Date
--------------------------------------------------------------------------------------------------------------------------------------------- 721. (THEORY and FEATURE) People Analytics - Handwriting Recognition - Inner Product Space of Contour Interpolated Polynomials - 17 October 2019 - related to 714,715,720 --------------------------------------------------------------------------------------------------------------------------------------------- 1.In continuation of the previous commit for drawing rasterized contour curves, handwriting contours are interpolated by SciPy.interpolate splprep() 2.Points on contour curves are stack()-ed to 2 dimensional arrays and contours having more than 3 points are interpolated by spline polynomials of degree less than number of points per contour. 3.PdfPages "pdf damaged" error in DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies.pdf has been resolved by PdfPages close() which flushes the buffer. 4.splprep (https://docs.scipy.org/doc/scipy-1.2.1/reference/generated/scipy.interpolate.splprep.html) creates B-Spline curves which are exact but less smooth as opposed to Bezier curves which are smoother, depend on control points and are approximate fit (passes through most of the points). 5.Boolean flag TopologicalRecognition=True/False has been introduced in __main__ to choose topological handwriting recognition instead of Convolution Neural Network Pattern Recognition. 6.Logs and Pdf images for contour polynomials are at: testlogs/DeepLearning_ConvolutionNetwork_BackPropagation.log.17October2019 testlogs/DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies.pdf 7.By defining an inner product space of contour polynomials <f,g>=DefiniteIntegral(f(x)g(x)dx), distance function is obtained for verifying if 2 contours belonging to different handwritten alphanumerics are "similar" or "orthogonal" (polynomials f and g are orthogonal if <f,g>=0). This is an algebraic but less topological similarity.
Srinivasan Kannan
More... about 7 years ago
(THEORY and FEATURE) People Analytics - Handwriting Recognition - Contour Homotopies - Matplotlib Rasterization
Srinivasan Kannan
More... about 7 years ago
---------------------------------------------------------------------------------------------------------------------------------------------- 720. (THEORY and FEATURE) People Analytics - Handwriting Recognition - Contour Homotopies - Matplotlib Rasterization - 16 October 2019 - related to 714 and 715 ---------------------------------------------------------------------------------------------------------------------------------------------- 1.This commit updates handwriting_recognition() in DeepLearning_ConvolutionNetwork_BackPropagation.py to additionally raster the contours extracted from the handwritten alphanumeric. 2.Rasterization presently invokes matplotlib plot() having parameter rasterized=True and presets dots-per-inch parameter to 100 in figure() for restricting number of pixels. 3.Contours from each recognized written alphanumeric are looped and plotted by matplotlib plot() setting rasterized=True. 4.There are better alternatives to this primitive rastering - rasterio (https://rasterio.readthedocs.io/en/stable/) and wavelet-rasterization (https://github.com/ufoym/wavelet-rasterization/) and desired rasterization is similar to illustrations in https://github.com/ufoym/wavelet-rasterization which is a coarse rectification required for hyperbolic arc in Computational Geometric Factorization (But there is a windows DLL build dependency). Matplotlib dots-per-inch and rasterized=True creates fine grained rasterized curves. 5.Each contour is approximable by a Bezier-like curve (https://en.wikipedia.org/wiki/B%C3%A9zier_curve) which is an instance of homotopy because bezier curves have endpoints/control points unchanged and intermediate points on curve are puppet-animated (most applicable in Graphics). 6.Handwriting can be simulated by a Bezier curve having unchanged endpoints or control points which is a homotopic continuous deformation. 7.Original handwriting is reconstructible by homotopy and pasting lemma - by pasting together polynomials for individual contours. 8.Logs and images for this commit are at testlogs/ (handwritten 1(style 1) and 1(style 2), 1 and 8 are contour-rasterized and compared): DeepLearning_ConvolutionNetwork_BackPropagation.log.tar.xz DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies_1_1.jpg DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies_1_2.jpg DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies_8_1.jpg DeepLearning_ConvolutionNetwork_BackPropagation.Raster.Homotopies.pdf 9.Every contour can be fit to a bezier polynomial and two alphanumerics have to be verified if they are homotopic (has a dependency on topology libraries). This is non-trivial because every homotopically "similar" contour in two alphanumerics have to be pairwise compared and number of contours per alphanumeric could be in hundreds.
Srinivasan Kannan
More... about 7 years ago
Theory - Computational Geometric Factorization - PRAM Rasterization of
Srinivasan Kannan
More... about 7 years ago
Theory - Computational Geometric Factorization - PRAM Rasterization of hyperbolic arc on pixel grid
Srinivasan Kannan
More... about 7 years ago
(THEORY) Complement Diophantines over Integers (Z), Polynomial Interpolation, Polynomial Reconstruction Problem, Error Correcting Codes, MRDP theorem, Lagranges Four Square Theorem and Decidability of Complementation, Ramsey 2-coloring of sequences, Set Partitions to Tile Cover reduction, Factorization, Lagrangian tiling of 2-dimensional space, ZF and Axiom of Choice (AOC) and Complement Choice functions, Depth-2 majority voting social choice circuits - an alternative interpretation based on diagonalization - updated for Complement Diophantines over Integers
Srinivasan Kannan
More... about 7 years ago
(THEORY) Complement Diophantines, Polynomial Interpolation, Polynomial Reconstruction Problem, Error Correcting Codes, MRDP theorem, Lagranges Four Square Theorem and Decidability of Complementation, Ramsey 2-coloring of sequences, Set Partitions to Tile Cover reduction, Factorization, Lagrangian tiling of 2-dimensional space, ZF and Axiom of Choice (AOC) and Complement Choice functions, Depth-2 majority voting social choice circuits - an alternative interpretation based on diagonalization - Correction
Srinivasan Kannan
More... about 7 years ago
719. (THEORY) Complement Diophantines, Polynomial Interpolation, Polynomial Reconstruction Problem, Error Correcting Codes, MRDP theorem, Lagranges Four Square Theorem and Decidability of Complementation, Ramsey 2-coloring of sequences, Set Partitions to Tile Cover reduction, Factorization, Lagrangian tiling of 2-dimensional space, ZF and Axiom of Choice (AOC) and Complement Choice functions, Depth-2 majority voting social choice circuits - an alternative interpretation based on diagonalization
Srinivasan Kannan
More... about 7 years ago
Theory - Exact Rasterization of Algebraic Curves - Computational
Srinivasan Kannan
More... about 7 years ago
(THEORY and FEATURE) People Analytics - Set Partition Analytics based Drone Electronic Voting Machine - updated
Srinivasan Kannan
More... about 7 years ago
--------------------------------------------------------------------------------------------------------------------------------------------- 718. (THEORY and FEATURE) People Analytics - Set Partition Analytics based Drone Electronic Voting Machine - updated - 9 October 2019 - related to 713 --------------------------------------------------------------------------------------------------------------------------------------------- 1.Streaming_SetPartitionAnalytics.py has been changed to update electronic_voting_machine() function - New semaphore mutual exclusion synchronized primitive block (BoundedSemaphore) has been added to guard the voting critical section. Maximum value of semaphore is set to 1 voter. 2.Voting code has been updated to check if voter has already voted by Voted list and skip voting if yes. Voter id(s) are appended to Voted array after voting within synchronized block. 3.Additional encryption has been introduced by invoking SHA256 Passlib encrypt on the ripemd160 publicuniqueid hex hash digest and append the encrypted voter id to EVM bucket. 4.This EVM assumes the function electronic_voting_machine() is invoked from a Drone python code which should have already navigated to the voter address and done the user/password authentication by smartcard. 5.Single Signon username/password authentication has already been implemented in NeuronRain GUI - python-src/webserver_rest_ui/NeuronRain_REST_WebServer.py - by passlib OAuth having MongoDB and Redis credential storage - which might as well be used as alternative caller for this EVM. 6.logs for preventing repetitive voting and SHA256 buckets are at testlogs/Streaming_SetPartitionAnalytics.EVM.log.9October2019 7.This Set Partition EVM is also a Dynamic Graph Partition (Similar Worst Case Execution Time - Survival Index - Dynamic Graph Partition OS Scheduler is described in https://github.com/shrinivaasanka/Grafit/blob/master/course_material/NeuronRain/AdvancedComputerScienceAndMachineLearning/AdvancedComputerScienceAndMachineLearning.txt - Hashing Dynamic Sets) because voting is dynamic and voters related among themselves might vote for different candidates which creates edges of dynamic people partition graph (voter vertices related by edges are partitioned into candidate buckets) 8.As mentioned in previous section, Left Inverse of Voter Bucket Objects is the Candidate voted for connecting peer voters by an equivalence relation.
Srinivasan Kannan
More... about 7 years ago
(THEORY) Computational Geometric Factorization - Parallel Planar Point Location of factors in Monotone Subdivision formed by Hyperbolic Arc Pixel Polygons with no necessity for Ray Shooting Queries, Van Der Waerden Ramsey Coloring of Sequences - Rectification of Hyperbolic Arc by Line segments (not necessarily axis parallel) - References (Bresenham-like rectification of hyperbolic arc is axis parallel)
Srinivasan Kannan
More... about 7 years ago
Theory - Majority Hardness Amplification Lemma, Noise Sensitivity, Sensitivity, Block Sensitivity, Condorcet Jury Theorem
Srinivasan Kannan
More... about 7 years ago
Theory - Topological Face Recognition - Shape Grammar for Face Recognition
Srinivasan Kannan
More... about 7 years ago
(THEORY) Cognitive Element in the context of Category theory, Factorization and Set Partitions
Srinivasan Kannan
More... about 7 years ago
(*) (THEORY) People Analytics - Topological Face Recognition - Product Homotopies (*) (THEORY and FEATURE) Economic Intrinsic Merit - Gravity Model of Volume of Trade and GDP as fitness measure
Srinivasan Kannan
More... about 7 years ago
------------------------------------------------------------------------------------------------------------------------------------------- 716. (THEORY and FEATURE) Economic Intrinsic Merit - Gravity Model of Volume of Trade and GDP as fitness measure - 2 October 2019 ------------------------------------------------------------------------------------------------------------------------------------------- 1.Intrinsic Fitness or Intrinsic Merit has been traditionally applied to denote innate ability of a vertex in Social Networks. Fitness model is applicable in Economic Networks too e.g International Trading Network(ITN), Production Networks(https://economics.mit.edu/files/9790). 2.https://arxiv.org/pdf/1409.6649.pdf discusses various measures of Economic fitness including GDP and quantifying volume of trade between nations - Quoted excerpts: 2.1 "...The standard model of non-zero trade flows, inferring the volume of bilateral trade between any two countries from the knowledge of their Gross Domestic Product (GDP) and mutual geographic distance (D), is the so-called ‘gravity model’ of trade [21–25]. In its simplest form, the gravity model predicts that the volume of trade between countries i and j is Fij = α GDP^βi · GDP^βj / Dij^k..." 2.2 "...where zi ≡ e^−θi and pij ≡ zi*zj/(1+zi*zj). The latter represents the probability of forming a link between nodes i and j, which is also the expected value <aij> = zi*zj/(1 + zi*zj)= pij . (4) ..." 2.3 "...It should be noted that eq.(4) can be thought of as a particular case of the so-called Fitness Model [41], which is a popular model of binary networks where the connection probability pij is assumed to be a function of the values of some ‘fitness’ characterizing each vertex. Indeed, the variables ~z∗ can be treated as fitness parameters [6, 37] which control the probability of forming a link. A very interesting correlation between a fitness parameter of a country (assigned by the model) and the GDP of the same country was found [37]. This relation is replicated here in Fig. 2, where the rescaled GDP of each country (gi ≡ GDPi/sum(GDPi)) is compared to the value of the fitness parameter z∗i obtained by solving eq.(5). The red line is a linear fit of the type zi = √a · gi...." 3. This commit defines a new function EconomicMerit() in LinearAndLogisticRegression.py which computes trade link probabilities and gravity model between countries based on GDP as intrinsic fitness measure and distances between countries. Reference 694.1 defines GDP fitness as a linear regression dependent on macroeconomic variables.
Srinivasan Kannan
More... about 7 years ago
Theory and Feature - Reduction from Set Partition to Exact Square Tiles Cover of Rectangular Region, Chinese Remainder Theorem for reduction to arbitrary dimensional space, Lagrange Four Square Theorem, Complementary Sets, Computational Geometric Factorization, Space Filling, Tile Cover, Packing, Set Partition/Histogram Analytics, LSH Partitions, Separate Chaining Hashtables
Srinivasan Kannan
More... about 7 years ago
(THEORY and FEATURE) Topological Handwriting Recognition - Product Homotopies
Srinivasan Kannan
More... about 7 years ago
Theory - Rectification of Hyperbolic Arc and Bresenham algorithm - Computational Geometric Factorization
Srinivasan Kannan
More... about 7 years ago
Theory - LSVR - Tensor decomposition of Video Tensor
Srinivasan Kannan
More... about 7 years ago
(THEORY and FEATURE) People Analytics - Set Partition based Electronic Voting Machine implementation and a NeuronRainApps Drone usecase
Srinivasan Kannan
More... about 7 years ago
------------------------------------------------------------------------------------------------------------------------------------------- 713. (THEORY and FEATURE) People Analytics - Set Partition based Electronic Voting Machine implementation and a NeuronRainApps Drone usecase - 24 September 2019 - related to 620,648 ------------------------------------------------------------------------------------------------------------------------------------------- Voter Received Encrypted Paper Audit Trail (VREPAT) Theoretical Voting Machine based on Set Partition Bucketing described previously is partially implemented as a futuristic drone usecase (Autonomous Delivery Usecases - https://syncedreview.com/2019/08/24/autonomous-delivery-moves-from-research-labs-to-the-streets/) in this commit - this is not a full-fledged implementation but conceptual: (*) Usual state-issued unique id(s) are singletons stored in centralized cloud and vulnerable to breaches/fakes creating identity disputes. (*) To redress this an alternative dual unique id (public id and private id) similar to public-key-infrastructure is assumed which is as strong as RSA and Diffie-Hellman protocols (*) Public unique id is a hash digest of a string which uniquely identifies a person e.g concatenation of <VoterID><Full Name><Parent/Spouse Name><Permanent Address> is always a unique public id. Private id is a password for a voter issued along with a Public unique id. (*) Polling Station is a Drone - it contains a set partition bucketing voting machine which based on addresses or public unique id(s) in the voter list automatically navigates to an address and lands on a voter doorstep. Voter casts h(is|er) vote on the drone booth by public unique id card and private password and receives encrypted receipt of vote. (*) This Dynamic Drone Polling Booth is inspired by Autonomous Online Shopping Delivery and Mobile ATMs available already (which work on smartcard and password authentication) and has many advantages over traditional static polling stations which satisfy constraints of John Hopkins Paper to some extent: (*) Voting need not be limited to a day and can sprawl multiple days. (*) Voter can do advanced booking for drone and it lands on demand making the voting behaviour difficult to predict. (*) All households are covered and thus maximizes polling percentage (*) Completely automated election - Minimizes or obviates polling stations, cost, queueing and personnel and thus human tamper-proof - requires one drone per polling precinct of few hundred voters (*) Ensures privacy (*) Reduces malpractices because of no crowding, each voter independently votes, polling happens at doorstep at time of choosing which is unpredictable and unique demographic/geographic identity is electronically verified by drone navigation to unique address minimizing bogus voting (no voter is denied voting because it could not have been already cast in h(is|er) name - GPS drone navigation is an additional geographic uniqueness). (*) This commit defines a new minimal voting function electronic_voting_machine() in Streaming_SetPartitionAnalytics.py which accepts voted_for, example voter id card contexts of the author from testlogs/Streaming_SetPartitionAnalytics_EVM/ (no text parsing is done) and makes ripemd160 hash digests from them. EVM is a dictionary set partition object mapping candidate voted_for to list of voters' public uniqueid hash digests - does not use counters and private id authentication. (*) Complexity theoretically, drone voting is reducible to Motion planning (A-Star algorithm) and travelling salesman problem (TSP) because drone has to efficiently navigate the voter vertices spread across geographically (e.g hamiltonian) and thus NP-complete.
Srinivasan Kannan
More... about 7 years ago
------------------------------------------------------------------------------------------------------------------------------------------- 713. (THEORY and FEATURE) People Analytics - Set Partition based Electronic Voting Machine implementation and a NeuronRainApps Drone usecase - 24 September 2019 - related to 620,648 ------------------------------------------------------------------------------------------------------------------------------------------- Voter Received Encrypted Paper Audit Trail (VREPAT) Theoretical Voting Machine based on Set Partition Bucketing described previously is partially implemented as a futuristic drone usecase (Autonomous Delivery Usecases - https://syncedreview.com/2019/08/24/autonomous-delivery-moves-from-research-labs-to-the-streets/) in this commit - this is not a full-fledged implementation but conceptual: (*) Usual state-issued unique id(s) are singletons stored in centralized cloud and vulnerable to breaches/fakes creating identity disputes. (*) To redress this an alternative dual unique id (public id and private id) similar to public-key-infrastructure is assumed which is as strong as RSA and Diffie-Hellman protocols (*) Public unique id is a hash digest of a string which uniquely identifies a person e.g concatenation of <VoterID><Full Name><Parent/Spouse Name><Permanent Address> is always a unique public id. Private id is a password for a voter issued along with a Public unique id. (*) Polling Station is a Drone - it contains a set partition bucketing voting machine which based on addresses or public unique id(s) in the voter list automatically navigates to an address and lands on a voter doorstep. Voter casts h(is|er) vote on the drone booth by public unique id card and private password and receives encrypted receipt of vote. (*) This Dynamic Drone Polling Booth is inspired by Autonomous Online Shopping Delivery and Mobile ATMs available already (which work on smartcard and password authentication) and has many advantages over traditional static polling stations which satisfy constraints of John Hopkins Paper to some extent: (*) Voting need not be limited to a day and can sprawl multiple days. (*) Voter can do advanced booking for drone and it lands on demand making the voting behaviour difficult to predict. (*) All households are covered and thus maximizes polling percentage (*) Completely automated election - Minimizes or obviates polling stations, cost, queueing and personnel and thus human tamper-proof - requires one drone per polling precinct of few hundred voters (*) Ensures privacy (*) Reduces malpractices because of no crowding, each voter independently votes, polling happens at doorstep at time of choosing which is unpredictable and unique demographic/geographic identity is electronically verified by drone navigation to unique address minimizing bogus voting (no voter is denied voting because it could not have been already cast in h(is|er) name - GPS drone navigation is an additional geographic uniqueness). (*) This commit defines a new minimal voting function electronic_voting_machine() in Streaming_SetPartitionAnalytics.py which accepts voted_for, example voter id card contexts of the author from testlogs/Streaming_SetPartitionAnalytics_EVM/ (no text parsing is done) and makes ripemd160 hash digests from them. EVM is a dictionary set partition object mapping candidate voted_for to list of voters' public uniqueid hash digests - does not use counters and private id authentication. (*) Complexity theoretically, drone voting is reducible to Motion planning (A-Star algorithm) and travelling salesman problem (TSP) because drone has to efficiently navigate the voter vertices spread across geographically (e.g hamiltonian) and thus NP-complete.
Srinivasan Kannan
More... about 7 years ago
Computational Geometric Factorization - Parallel Planar Point Location of factors in Monotone Subdivision formed by Hyperbolic Arc Pixel Polygons with no necessity for Ray Shooting Queries, Van Der Waerden Ramsey Coloring of Sequence - Update on Union of Arithmetic Progressions and Generalized Planar Point Location
Srinivasan Kannan
More... about 7 years ago
(THEORY) Category Theory formulation of EventNet - Objects, Morphisms, Functors - Tensor Decomposition of EventNet Causality Functor Tensor
Srinivasan Kannan
More... about 7 years ago
(THEORY) Computational Geometric Factorization - Gamma Approximation of Sum of Binary Search times - Summation for first sqrt(N) tiles and why non-average case sequential factorization is difficult
Srinivasan Kannan
More... about 7 years ago
(FEATURE) Social Network Analysis - People Analytics - Contextual Name Parsing and Syllable Vector Compression of Strings
Srinivasan Kannan
More... about 7 years ago
----------------------------------------------------------------------------------------------------------------------------------------- 711. (FEATURE) Social Network Analysis - People Analytics - Contextual Name Parsing and Syllable Vector Compression of Strings - 18 September 2019 ----------------------------------------------------------------------------------------------------------------------------------------- 1.Regex Pattern Matching in nameparser() has been changed to ignore case by flag re.IGNORECASE 2.Few more ID context texts are being regex matched in nameparser() which are parsed texts and OCRs (by pdftotext and gocr) from pdf and image IDs of the author uploaded in https://sourceforge.net/projects/acadpdrafts/files/?source=navbar - Public Distribution System ID card linked to Unique Aadhaar IDs and Permanent Account Number (PAN) which is Aadhaar linked - https://sourceforge.net/projects/acadpdrafts/files/NewRationCard2.pdf/download, https://sourceforge.net/projects/acadpdrafts/files/NewRationCard1.pdf/download and https://sourceforge.net/projects/acadpdrafts/files/PAN.jpg/download 3.These three ID cards are examples of Both Indian convention of Second name followed by First name and International standard of First name followed by Second name. https://sourceforge.net/projects/acadpdrafts/files/NewRationCard2.pdf/download, https://sourceforge.net/projects/acadpdrafts/files/NewRationCard1.pdf/download have Unique IDs linked to both formats but in different languages - Srinivasan Kannan கண்ணன சீனிவாசன 4.nameparser() is invoked on previous three ID contexts for full name "Kannan Srinivasan" and first and second names are printed 5.syllable_boundary_text_compression() from CompressedSensing implementation is invoked for 3 different spellings of same name "Shrinivaasan, Shrinivas, Srinivasan" and syllable vectors (and their vowelless versions) are printed: ============================================================================== Vowelless Syllable Vector Compression for text - Shrinivaasan : ([u'Shrini', u'vaasan'], u'Shr_n_-v__s_n') Vowelless Syllable Vector Compression for text - Shrinivas : ([u'Shrini', u'vas'], u'Shr_n_-v_s') Vowelless Syllable Vector Compression for text - Srinivasan : ([u'Srini', u'vasan'], u'Sr_n_-v_s_n') ====================================================================== 6.Distance between these syllable vectors has to be phonetic by finding syllable-wise audio similarity - every syllable is an audio waveform and two names are compared by syllable distances between vectors of waveforms e.g ["Shrini","vas"] and ["Srini","vasan"]. This has been omitted presently because syllable-to-audio tools are required. 7.Comparison to Match Rating Codex from JellyFish is also printed: Match ratings for same name of differing spellings - [Shrinivaasan,Shrinivas,Srinivasan]: [u'SHRVSN', u'SHRNVS', u'SRNVSN'] 8.logs for this commit and previous are at testlogs/SocialNetworkAnalysis_PeopleAnalytics.log.18September2019
Srinivasan Kannan
More... about 7 years ago
Updated Requirements.txt
Srinivasan Kannan
More... about 7 years ago