ml   39909

« earlier    

Debugging machine learning - Michał Łopuszyński - YouTube
Debugging machine learning apps is hard. I feel that this topic is important, however, relatively rarely touched compared to, e.g., latest models or new interesting ML applications. In this talk, I will try try to fill that gap by discussing best practices, recommendations and my own experience on the subject.
ML  video  intepretation 
22 hours ago by foodbaby
Classification is a fundamental problem in machine learning, data mining, and
computer vision. In practice, interpretability is a desirable property of classification models
(classifiers) in critical areas, such as security, medicine, and finance
ML  explanation  visualization 
23 hours ago by foodbaby
BIDViz: Real-time Monitoring and Debugging of Machine Learning Training Processes | EECS at UC Berkeley
Artificial Intelligence is a thriving field with many applications, whether in automating routinary human labors or support basic research such as diagnosing diseases (Goodfellow et. al., 2016). Deep learning, or machine learning with deep neural networks, is particularly successful in providing amazing human like intelligence, such as image captioning or translation, because it has a more general model of the world and makes relatively few assumptions on the world it trying to model (Goodfellow et. al., 2016). However, comparing to the tools for programming and debugging, the tooling for building and learning deep learning models is not as good.

In this report, we present BIDViz, a visualization platform that allows data scientists to visualize and debug his/her model interatively while it is training. BIDViz is applicable to general machine learning but is oriented toward deep neural models, which are often challenging to fully understand. BIDViz emphasizes dynamic exploration of models by allowing the execution of arbitrary metrics or commands on the model while it is training.
ML  training  debugging 
23 hours ago by foodbaby
[1703.04730] Understanding Black-box Predictions via Influence Functions
How can we explain the predictions of a black-box model? In this paper, we use influence functions -- a classic technique from robust statistics -- to trace a model's prediction through the learning algorithm and back to its training data, thereby identifying training points most responsible for a given prediction. To scale up influence functions to modern machine learning settings, we develop a simple, efficient implementation that requires only oracle access to gradients and Hessian-vector products. We show that even on non-convex and non-differentiable models where the theory breaks down, approximations to influence functions can still provide valuable information. On linear models and convolutional neural networks, we demonstrate that influence functions are useful for multiple purposes: understanding model behavior, debugging models, detecting dataset errors, and even creating visually-indistinguishable training-set attacks.
ML  explanation  interpretation 
23 hours ago by foodbaby
What the SATs Taught Us about Finding the Perfect Fit | Stitch Fix Technology – Multithreaded
On the Stitch Fix Algorithms team, we’ve always been in awe of what professional stylists are able to do, especially when it comes to knowing a customer’s si...
algorithm  fashion  statistics  favorite  machinelearning  ml 
yesterday by morganwatch
Favourites 2017: KDnuggets - The Data Science Puzzle, Explained
DeepLearning  BigData  ML  DL  MachineLearning  AI  from twitter_favs
yesterday by neuralmarket

« earlier    

related tags

101  2017  abtesting  advances  ai  algorithm  algorithms  analytics  ar  archive  art  article  artificalintelligence  artificialintelligence  aws  bandit  bandits  beginner  bigdata  blockchain  blog  blogger  bots  cheatsheet  classification  code  control  coreml  course  data  datascience  debugging  deep-learning  deeplearning  descent  discussion  dl  education  energy  experience  explanation  extremely_useful  fashion  favorite  finance  fintech  food  framework  ftrsn_posted  games  generativemodels  golang  google  gp  gradient  graph  graphics  guide  health  healthtech  hedgefund  hn  hogg  indexing  intepretation  interpretation  iot  jobs  jupyter  kubernetes  language  languages  learning  library  machine-learning  machine  machinelearning  marketing  martech  math  mobile  model  mstechsummit  netflix  network  neuralnetworks  nlp  nutrition  obama  optimization  pdf  personalization  personalized  pipeline  ppt  pptx  presentation  production  programming  python  r  readthis  recommendation  recommendations  reference  regression  research  resources  science  set  slide  slides  social  socialmedia  speech  startups  statistics  tech  technology  tensorflow  testing  toolkit  toread  toronto  training  tutorial  tutorials  unity  usa  validation  version  video  vision  visualization  vr 

Copy this bookmark: