You want to get updates? Please join Data Sci­ence & Machine Learn­ing Newslet­ter Linked Group

  • Ulti­mate Guide To Sta­tis­ti­cal Sig­nif­i­cance Tests in R
    • Imple­ment and inter­pret the com­monly used sta­tis­ti­cal sig­nif­i­cance tests in R, the pur­pose, when to use and how to inter­pret the result.
  • Word2Vec (skip-gram model)
    • The skip-gram neural net­work model is actu­ally sur­pris­ingly sim­ple in its most basic form. Train a sim­ple neural net­work with a sin­gle hid­den layer to per­form a cer­tain task, but then we’re not actu­ally going to use that neural net­work for the task we trained it on! Instead, the goal is actu­ally just to learn the weights of the hid­den layer–we’ll see that these weights are actu­ally the “word vec­tors” that we’re try­ing to learn.
    • Part 1
    • Part 2
  • Arti­fi­cial intel­li­gence sug­gests recipes based on food pho­tos
    • Researchers from MIT’s Com­puter Sci­ence and Arti­fi­cial Intel­li­gence Lab­o­ra­tory (CSAIL) believe that ana­lyz­ing pho­tos like these could help us learn recipes and bet­ter under­stand people’s eat­ing habits. In a new paper with the Qatar Com­put­ing Research Insti­tute (QCRI), the team trained an arti­fi­cial intel­li­gence sys­tem called Pic2Recipe to look at a photo of food and be able to pre­dict the ingre­di­ents and sug­gest sim­i­lar recipes.
  • Com­par­ing Dif­fer­ent Species of Cross-Validation

    • An all in one Python3 Data Sci­ence Pack­age. Easy visu­al­i­sa­tion, data min­ing, data prepa­ra­tion and machine learning.
  • DARLA: Improv­ing Zero-Shot Trans­fer in Rein­force­ment Learn­ing
    • Domain adap­ta­tion is an impor­tant open prob­lem in deep rein­force­ment learn­ing (RL). In many sce­nar­ios of inter­est data is hard to obtain, so agents may learn a source pol­icy in a set­ting where data is read­ily avail­able, with the hope that it gen­er­alises well to the tar­get domain. We pro­pose a new multi-stage RL agent, DARLA (Dis­en­tAn­gled Rep­re­sen­ta­tion Learn­ing Agent), which learns to see before learn­ing to act.
  • How to squeeze the most from your train­ing data
    • In many cases, get­ting enough well-labelled train­ing data is a huge hur­dle for devel­op­ing accu­rate pre­dic­tion sys­tems. Here is an inno­v­a­tive approach which uses SVM to get the most from train­ing data.
  • The Truth About Bayesian Pri­ors and Over­fit­ting
    • Many of the con­sid­er­a­tions we will run through will be directly applic­a­ble to your every­day life of apply­ing Bayesian meth­ods to your spe­cific domain.