The Technology

  • Decentralized, Privacy-Preserving ML

    A proprietary federated learning environment enables unprecedented levels or privacy

  • Eliminate Traditional Model Training

    Automated model & feature selection capabilities

  • Rapid Model

    No rescanning of data required

  • Continuous
    Retraining Models

    Reduces model maintenance

  • Significantly Lower Compute Costs

    SymetryML brings unmatched efficiency to machine learning

Unique ML Capabilities

  • AutoML - feature engineering & model selection
  • Incremental/Decremental Learning
  • Quick Iterative Prototyping
  • Add / Remove Variables On-the-fly
  • Real-time Data Exploration Dashboard
  • 'Forget' Functionality
  • Streaming Anomaly Detection

Scalable & Easy To Use

  • Low Code Graphical UI or Rest API (JSON)
  • High Performance Computing & GPU Quick Model Deployment
  • Parallel Computing Enabled
  • Leverage Existing Hardware with Multi-Processor & Multi-Core Architecture
  • Integrations with Apache Spark & Apache Kafka
  • Amazon or Other Cloud Architectures
  • Simple & Flexible Pricing Structures (usage, seats, unlimited)
BI & Data Visualization
  • Continuous Data Streaming
  • Distributed Computing
Data Connections

SymetryML’s Proprietary Data Structure

SymetryML’s continuous machine learning techniques can update a trained model when each new data point occurs. This allows a model to not only predict, but learn in real-time as well. In other words, there’s no need to store a historical training set and retrain a model periodically, the model can stay up-to-date.

To accomplish this, SymetryML uniquely decouples learning a model, from the actual data by extracting key statistical values and creating a Proprietary Statistical Representation (PSR) of your entire data set. This compression-like technique provides several unique benefits, including rapid model building, continuous retraining and lowers overall compute costs and resources.

Unique Machine Learning Techniques

  • Federated Learning

    Federated Learning, often referred to as "Collaborative Learning", is an emerging area of AI. It is a machine learning technique that trains an algorithm across multiple decentralized datasets, edge devices or servers. The actual, raw data is never seen, shared or moved, instead only "summarized learnings" of models (or in SymetryML's case - the PSR) are shared amongst members of a federation through SymetryML's encrypted solution. This allows more data to be leveraged in a privacy friendly way, which means knowledge and insights can be shared without the hassle of anonymizing data or the security risk of aggregating sensitive data into a central server.

  • AutoML

    AutoML allows users to focus their analysis on a particular target variable. SymetryML will automatically remove irrelevant attributes, avoid issues with collinearity, filter outliers, and create interaction pairs amongst the remaining set of inputs.

  • AutoSelect

    Leveraging SymetryML's unique capabilities to build multiple models very quickly, SymetryML's AutoSelect feature allows users to rapidly build many different models using different model tasks as well as trying many combinations of input attributes, then using out-of-sample data to 'automatically select' the best model.

  • Quick Iterative Prototyping

    Unlike the vast majority of other data mining solutions, SymetryML separates the tasks of learning and model building into two discrete steps: (1) learning the data and (2) building the actual model. The classical approach assumes these two steps are synonymous by defining model building within the actual learning phase.

    SymetryML achieves the separation by scanning the input data in such a way as to defer model creation to a later stage. This allows SymetryML to reduce the memory footprint required to learn large data sets and allows models to be created on the fly, resulting in more experimentation and quicker deployments.

    For business applications, this translates into more robust results, faster time to market and the potential for higher levels of confidence in the accuracy of the models themselves.

  • Incremental / Decremental Learning

    Traditional analytical toolsets require full scan of the historical data to update their internal models, which often causes models to become outdated quickly after they are deployed. SymetryML avoids this problem by allowing incremental learning in real time utilizing the latest data points available.

  • Add / Remove Variables On-the-Fly

    SymetryML has the unique capability of adding a new attribute (variable) on the fly, without the necessity of pooling new data with old data or immediately discontinuing the use of an attribute identified as adversely affecting model performance.

  • Real Time Exploration of Data

    As SymetryML incorporates and 'learns' new data, it updates its internal representation of the data with near zero-latency. This patented technique allows for real time exploration of the data using statistical tools without the need for a complete table scan that would be the case in traditional data mining, thus allowing the user to process more data at a faster rate than any other technology available today.

  • 'Forget' Functionality

    SymetryML has the unique functionality to 'forget' data. When SymetryML learns from new data, it updates an internal representation of the overall data, effectively allowing the system to forget specific parts based on what was learned afterward.

  • Quick Model Deployment

    Use the SymetryML-Web Application to explore, analyze your data and create predictive models. Models built within SymetryML-Web are automatically accessible via the SymetryML-REST API enabling quick integration with your existing data flows.

Easy To Implement & Scale

  • Integration with Apache Spark Cluster

    SymetryML can leverage Apache Spark clusters for distributed learning. If data can be entered into Spark in any of the following Spark data structures (RDD or DataSet), then SymetryML can process it in a distributed fashion.

  • Integration with Apache Kafka

    SymetryML can stream data from Apache Kafka clusters. Whether Kafka topics use single or multiple partitions – with tons of I/O – SymetryML can scale to process them in real-time allowing users to:

    • Build streaming predictive machine learning solution that adapt in real time
    • Inspect various statistics about your streaming data in real time using the Exploration API
  • High Performance Computing and GPU

    There are various definitions of Big Data: 3V (Volumes, Variety, Velocity), 4V (3V + Value). But one detail that analysts sometimes forget to mention is that data volume has 2 dimensions: (1) the number of records (2) the number of attributes (or fields/features) in each record. Data mining 10,000 records each with 100 attributes is a different problem than data mining 100 records each with 10,000 attributes. Finding relationships between and among 100 attributes is an easier task than finding such meaningful relationships between 10,000 attributes. SymetryML leverages the power of GPUs to tackle such broad data sets in a fraction of the time it takes other systems.

  • Parallel Computing Enabled

    Leverage the performance of modern-day parallel processors, like multi-core CPUs and GPUs, for increased performance during model rebuilds.

  • Leverage Existing Hardware with Multi-Processor and Multi-Core Architecture

    There is a paradox that is developing in the IT world: The price of hardware is dropping, but IT departments are spending more money because existing software is not fully utilizing the new parallel CPU architecture. SymetryML is a true parallel software, and it will automatically use all available computing resources on the host computer in order to perform its data mining tasks.


Schedule a demo