machine-learning   21543

« earlier    

[1801.05895] Sparsely Connected Convolutional Networks
Residual learning with skip connections permits training ultra-deep neural networks and obtains superb performance. Building in this direction, DenseNets proposed a dense connection structure where each layer is directly connected to all of its predecessors. The densely connected structure leads to better information flow and feature reuse. However, the overly dense skip connections also bring about the problems of potential risk of overfitting, parameter redundancy and large memory consumption. In this work, we analyze the feature aggregation patterns of ResNets and DenseNets under a uniform aggregation view framework. We show that both structures densely gather features from previous layers in the network but combine them in their respective ways: summation (ResNets) or concatenation (DenseNets). We compare the strengths and drawbacks of these two aggregation methods and analyze their potential effects on the networks' performance. Based on our analysis, we propose a new structure named SparseNets which achieves better performance with fewer parameters than DenseNets and ResNets.
neural-networks  binge-purge-cycles  architecture  algorithms  machine-learning  representation  to-write-about 
7 hours ago by Vaguery
slundberg/shap: Explain the output of any machine learning model using expectations and Shapley values.
shap - Explain the output of any machine learning model using expectations and Shapley values.
yesterday by hschilling

« earlier    

related tags

2018  :/  abstraction  acm  addresses  adversarial  ai  algorithms  api  apple  architecture  art  articles  artificialintelligence  arxiv  autoencoders  aws  binge-purge-cycles  biology  blog  books  by:yoshuabengio  classification  cloud  cnns  composition-decomposition  computer-vision  counterexample  courses  culture  cz  data-analysis  data-science  datetime  deep-learning  deeplearning  design  dl-example  dl-math  economy  error  event  evolutionary-algorithms  explanans  features  fourier  gans  generalization  generative-adversarial-network  geo  google-brain  google  gpu  graphics  h2o  hmm  ia  image-processing  infrastructure  iosdev  iot  javascript  jupyter  kaggle  keras  kubernetes  learning  lower-bounds  lstm  machinelearning  maps  medicine  ml  mltheory  mobile  model-class  neural-networks  neuralnetworks  nibble  nlp  openai  optimization  org:mat  pandas  paper  papers  performance  perturbation  philosophy  physics  prague  preprint  python  pytorch  read2of  regression  reinforcement-learning  representation  rnns  robotics  robust  saas  scalability  science  scikit-learn  search  security  self-driving-car  service  simulations  six@six  speedometer  state-of-art  statistical-arbitrage  statistics  structure  study-group  tensor-flow  tensorflow  text  theory  time-series  timezone  to-write-about  torch  trading-systems  trading  tutorial  uber  unity  universalism-particularism  voice  waves  webdev 

Copy this bookmark: