Deep Learning and Machine Learning

* BidMach

Description:

GPU-accelerated classical machine learning library

Features:

Logistic regression, SVM, LDA, SFA, NMF, ICA, random forests, clustering, word2vec

Multi GPU support:

Yes

Caffe

Description:

The Caffe deep learning framework makes implementing state-of-the-art deep learning easy.

Features:

Process over 40M images per day with a single NVIDIA K40 or Titan GPU.

Multi GPU support:

Single only

Caffe* Parallel

Description:

This is a faster framework for deep learning, it’s forked from BVLC/caffe (master branch). This allows data-parallel via MPI.

Features:

Using the GPU cluster processing mass image data

Multi GPU support:

Yes

Clarifai

Description:

Clarifai brings a new level of understanding to visual content through deep learning technologies. Clarifai uses GPUs to train large neural networks to solve practical problems in advertising, media, and search across a wide variety of industries.

Features:

GPU-based training and inference. Recognizes and indexes images with predefined classifiers, or with custom classifiers.

Multi GPU support:

Yes

Chainer

Description:

DL framework that makes the construction of neural networks (NN) flexible and intuitive.

Features:

Dynamic NN construction, which makes debugging easier. CPU/GPU-agnostic coding, which is promoted by CuPy, partially NumPy-compatible multidimensional array library for CUDA. Data-dependent NN construction, which fully exploits the control flows of Python without magic.

Multi GPU support:

Yes

* CNTK

Description:

Microsoft’s Computational Network Toolkit (CNTK) is a unified computational network framework that describes deep neural networks as a series of computational steps via a directed graph.

Features:

Supports many applications, including Speech Recognition, Machine Translation, Image Recognition, Image Captioning, Text Processing and Relevance, Language Understanding, Language Modeling

Multi GPU support:

Yes

Deeplearning4j

Description:

Deeplearning4j is the most popular deep learning framework for the JVM, and includes all major neural nets such as convolutional, recurrent (LSTMs) and feedforward.

Features:

Integrates with Hadoop and Spark to run distributed. Java and Scala APIs. Composable framework that facilitates building your own nets. Includes ND4J, the Numpy for Java.

Multi GPU support:

Yes

Dextro

Description:

Dextro’s API uses deep learning systems to analyze and categorize videos in real-time.

Features:

Object and scene detection, Machine transcription for audio Motion and movement detection.

Multi GPU support:

Yes

* IntelligentVoice

Description:

Far more than a transcription tool, this speech recognition software learns what is important in a telephone call, extracts information and stores a visual representation of phone calls to be combined with text/instant messaging and E-mail. Intelligent Voice’s search and alert makes it possible to tackle issues before they arise, address data security concerns and monitor physical access to data.

Features:

Advanced Speech Recognition across large data sets, JumpTo Technology, for data visualisation, E-Discovery, extraction from phone calls, IM & Email defining key phrases and emotional analysis. Compliance, defining key conversations and interactions

Multi GPU support:

Yes

Labellio

Description:

The world’s easiest deep learning web service for computer vision, which allows everyone to build own image classifier with only web browser.

Features:

Neural net fine-tuning for image data, data crawling, data browsing as well as drag-and-drop style data cleansing backed by AI support.

Multi GPU support:

Yes

* MatConvNet

Description:

CNNs for MathWorks MATLAB, allows you to use MATLAB GPU support natively rather than writing your own CUDA code

Features:

Building Blocks, Simple CNN wrapper, DagNN wrapper, cuDNN implemented

Multi GPU support:

Yes

MetaMind

Description:

Provides a deep learning API for image recognition and text sentiment analysis. Uses either prebuilt, public, or custom classifiers.

Features:

GPU-based training and inference. Recognizes image and analyzes text, creates and trains classifiers with tooling for uploading and managing datasets.

Multi GPU support:

Yes

* Neon

Description:

Neon is a fast, scalable, easy-to-use Python based deep learning framework that has been optimized down to the assembler level. Neon features a rich set of example and pre-trained models for image, video, text, deep reinforcement learning and speech applications.

Features:

Training, inference and deployment of deep learning models. Process over 442M images per day on a Titan X

Multi GPU support:

Yes

Theano

Description:

Theano is a symbolic expression compiler that powers large-scale computationally intensive scientific investigations.

Features:

Abstract expression graphs for transparent GPU acceleration.

Multi GPU support:

Single only

* Tensorflow

Description:

Google’s TensorFlow is an open source software library for numerical computation using data flow graphs. Nodes in the graph represent mathematical operations, while the graph edges represent the multidimensional data arrays (tensors) communicated between them.

Features:

TensorFlow is flexible, portable and performant creating an open standard for exchanging research ideas and putting machine learning in products.

Multi GPU support:

Yes

Torch7

Description:

Torch7 is an interactive development environment for machine learning and computer vision.

Features:

Computational back-ends for multicore GPUs.

Multi GPU support:

Single only

Trakomatic OSense, Otrack

Description:

Video Analytics Solution for retail, supermarkets, shopping mall and banking.

Features:

People detection & tracking, Crowd density estimation, Gender classification and age estimation, Person re-identification.

Multi GPU support:

Yes