Contact Us

Chat with a deep learning engineer

Request a Free Consultation

We help implement, deploy, and extend your deep learning applications. Please contact us to learn more.


Do you integrate with Hadoop?

Yes. Deeplearning4j works as a YARN app, run and provisioned by Hadoop as a first-class citizen. We also work within Spark, and enable it to run on multiple GPUs. Finally, we can simply use Hadoop as a data source by integrating with HDFS.

Where is my data stored? What do you do with it?

Your data stays with you. Our algorithms can process it wherever you want: on a local file system, on HDFS, or on S3. You don’t need to share it with anyone.

Do you have killer apps?

Lots. Deep learning is a form of machine perception. In every field where our perceptions can be quantified, deep learning can create killer apps. These include image, face and expression recognition; speech to text and machine interaction; sentiment analysis for text; and predictions in time-series data.

What distinguishes deep learning from machine learning?

While machine learning is capable of high levels of accuracy, they are usually attained after long effort and based on vast amounts of data. Deep learning can produce significantly more accurate results than machine learning. The problem, until now, has been the amount of time it took to train deep learning nets. Skymind accelerated deep learning and made it practical by distributing it across multiple processing units.

On a technical level, deep-learning networks are distinguished from the more commonplace single-hidden-layer neural networks by their depth; that is, the number of node layers through which data is passed in a multistep process of pattern recognition. More than three layers, including input and output, is deep learning. Anything less is machine learning. The number of layers affects the complexity of the features that can be identified.

Is Skymind a software as a service?

No. Think of us like this: What Red Hat is to Linux, and what Cloudera is to Hadoop, Skymind is to Deeplearning4j. Linux, Hadoop and DL4J are open-source, but certain implementations may require the expertise of a third party. If you need help, please contact us through the form to the left.

What’s special about Skymind?

First, Skymind’s distribution of Deeplearning4j was written for the JVM, while most deep-learning projects are coded in Python or Lua. While Python is a great research tool, Java is the de facto programming language of most businesses. Skymind plays nice with the production stack those businesses already use, and is accessible to most Java programmers.

Second, Skymind is distributed. That means it can run on many processing units in the cloud at once, which increases how fast its neural nets learn data. Speed, and the time it took to train nets, were barriers to using deep learning for many businesses. That barrier has been considerably lowered.

How can you guarantee the accuracy of your results?

Accuracy is relative to the data that is fed into the software. No machine or human is 100% accurate. Skymind can match or surpass any state-of-the-art machine-learning algorithm, and it can do so while performing feature extraction for you, saving data scientists thousands of hours of repetitive work. While results will depend on the problem to be solved, we have attained 97% accuracy analyzing images and text.

How will you help us tune our models?

Deep learning has a lot of knobs to turn. Each of those knobs represents a parameter for the software, and those parameters change according to the nature of the data and the problems you aim to solve. Skymind gives weeklong corporate workshops to teach Java engineers and data scientists the ins and outs of deep learning at scale. Post workshop, we will also help you analyze your data, diagnose its problems, adjust the knobs in light of that data, and produce the fastest, most accurate results possible. Please contact us if you’re interested.