tensorflow
background

TensorFlow is nowadays the primary software tool of deep learning. It is an open-source artificial intelligence library that applies data flow graphs to build models. It allows developers to create large-scale neural networks with many layers. TensorFlow is mainly used for: Classification, Perception, Understanding, Discovering, Prediction and Creation.

business challenges

The team's task was to create a fast and effective way to prototype, train and release machine learning models for the speech-processing framework



value delivered

We've developed a fast and effective approach that quickly turns a prototype into a production-ready machine learning component. One of the main advantages of this approach is that most of the code written at the first stage can be transfered to further stages. There is no need to do codegeneration or translation to other languages, frameworks etc. Moreover, in this way you have access to multiple versions of your version and can easily switch between them.



approach

Firstly, a prototype is built on a local machine using Tensorflow with CPU and smaller-size models. It is good practice to build a Tensorflow from sources because it doubles the speed of CPU calculations. As soon as the pipeline is debugged, the data collection code is ready and a simple baseline is estimated, we transfer training to a bigger GPU machine and switch to a more complex model (AWS p2.xlarge fits perfectly). When the desired accuracy is reached, the model is packed and and uploaded to the models repository using artifactory.

For Tensorflow, it's recommended to reuse most of the training code in the model serving module. If you are OK with using gRPC for service communication, then Tensorflow Serving is a powerful tool.


you may interested in other

case studies

  • Predictive
    Analytics

    view details

    View more
  • Large Scale
    Analytics

    view details

    View more

our expertise in

AI technologies

data mining

PCA
K-means
Decision trees
Linear models
PageRank

digital signal processing

Digital filters
DTW

machine learning

Deep learning
Probabilistic graphical models
CART
ensembles
unsupervised sound segmentation
recurrent models
bayesian approach
probabilistic programming
hmm

image processing and
computer vision

alexnet
vgg
vae

natural language
processing

PCA
TF-IDF
LDA
SVM
Naive bayes
word2vec
attention models

are you ready to see your software project getting real?

contact us


about us

Hi, we are Sciforce - a company where the integration of various branches of science builds up a powerful force to create robust software solutions. Working at the intersection of Computer Science with other technical, natural and humanitarian sciences let us go beyond traditional IT services and become both technical and scientific forces to our customers.