01/

Meet Turium Algoreus

The environment is more technologically challenging than at any time in history due to knowledge silos and the turf wars they enable within an enterprise.
Winning against knowledge silos requires an overmatch not by breaking down your operations to treat each like a linear problem but, rather, via a brain-like system that provides a unified vision through data supremacy. Data supremacy is a comprehensively transformative process that produces the most qualitative, well-curated, highly contextual data to enable data literacy and real-time decision precision across the enterprise.

02/

An Inside Look

A Hybrid Intelligence (HI) system that combines the very different processing strengths of Artificial Intelligence and the human brain to result in a symbiotic powerful platform. Think of ALGOREUS as the virtual twins of distributed neurons in the brain, linking historically siloed, disconnected systems to power smarter, more informed operations.

03/

Get Started in 4 Steps - See Impact in Days

OBSERVE

Connect to your systems and start extracting data immediately.

DECIDE

Explore the relationships in your enterprise data, apply machine learning at scale relevant to your problem space.

ORIENT

Automatically build your pipeline and generate business objects at scale.

ACT

Monitor your data and turn it into immediate impact with out-of-the-box use-cases.

03/

ALGOREUS Orient

FABRIQ DATA CONNECTION

500+ data connectors, leveraging an extensible plugin-based paradigm

Flexible ingress topology, which can leverage agent-based, REST, JDBC, and other approaches

Easy-to-configure schedules, success criteria, and permission models

Multimodal (structured, unstructured, streaming, IoT, geospatial, etc.)

DATA TRANSFORMATION

Flexible architecture with bundled engines

Low-code / no-code transformation (drag and drop interface)

Treating Data like Code (versioning, branching, full change management)

Full provenance

PIPELINE ORCHESTRATION

Build system that is engine-agnostic

Intelligent refreshing / state-tracking across all pipelines

Seamless integration with Fabriq’s health monitoring

04/

05/

Data Intelligence

Data Visualisation

Visually display interesting statistical properties of your dataset and expose unexpected data quality issues like outliers, correlations or missing values.

Automatic Data Insights

Visual and text descriptions for automatically detected trends and insights including topics in text, correlations, and outliers.

Pre-processing Transformers

Automatically include custom data preparation as part of your final deployed machine learning pipeline.

Outlier Detection

Expose issues or irregularities in data with better accuracy delivered through various proprietary algorithms.

Dataset Splitting

Save time and improve validation with a variety of built-in splitting techniques, including splitting randomly, by time, with stratification and with full customization via live code.

Missing Value Handling

Produce higher accuracy and better generalisation with end to end support for missing values in all parts of the machine learning pipeline.

04/

05/

Feature Engineering

Automated Feature Engineering

Increase accuracy and ROI with our proprietary feature engineering that automatically extracts non-trivial statistical information from your data.

Feature Encoding

Convert mixed data types (numeric, categorical, text, image, date/time, etc.) in a single dataset for use by machine learning algorithms.

Feature Transformation

Apply your domain knowledge to refine automated feature engineering outputs with fully customizable Python recipes.

Automated validation and cross validation

Improve accuracy, robustness and generalisation with a multitude of proprietary validation techniques, statistical methods and moving windows.

Per-Feature Controls

Disable feature engineering and feature selection for certain columns in your dataset, and pass them as-is to the model to satisfy your compliance requirements.

Automated feature selection

Reduce model complexity, produce faster inference time and better model interpretability with a multitude of proprietary feature selection techniques that automatically select the most predictive features for your dataset.

06/

ALGOREUS Decide

THE CORE MULTI-LAYERED

ONTOLOGY

Contains the key semantics of your world (objects and relations)

Contains the key kinetics of your world (Functions, Actions)

Integrated monitoring, and extensibility with external systems

DECISION CAPTURE / ENTERPRISE WRITEBACK

Structured mechanisms for capturing data from end users, back into the multi-layered ontology

Native frameworks for propagating data capture to external systems

External system responses can be woven into multi-step workflows

Full provenance

OUT-OF-THE-BOX OBJECT EXPLORATION

Provides a secure, scalable, point-and-click view into the ontology

Provides a chart-based paradigm, allowing for (among many other workflows) the navigation of multi-dimensional, real-time streaming data

Map provides a geospatial canvas for exploring the ontology

Each of the “base” applications is replete with a widget library that is continuously updated

CODE WORKBOOKS

An integrated, end-to-end workbench for model construction (PySpark, R, SparkSQL)

Native, secure data access for model builders (dataset and ontology paradigms)

Integrated model training, health, and management services

Flexible deployment options, for use in operations (batch and inference)

EXTERNAL MODEL INTEGRATION

Build and train your models in any industry-standard toolset

API-driven connectivity to the Ontology from those external tools

Promote into production through Algoreus, when ready

MODEL OBJECTIVES

“Mission Control” for models being used throughout Algoreus workflows

Rich, competitive evaluation of models; comparing performance

Binding directly to the Ontology, which provides a “type system” for models - allowing them to be leveraged in myriad operational settings

04/

05/

Automated Machine Learning
(AutoML)

AutoML is pervasive across the entire Platform: Powering everything from feature transformation to model selection, monitoring and deployment, robust autoML capabilities are the engine behind our ability to deliver AI.

Hyperparameter Autotuning

Increase accuracy, ROI and time savings with optimization across all components of the machine learning modelling pipeline delivered through a mix of our proprietary, genetic algorithm, Monte Carlo, Particle Swarm and Bayesian methods.

Champion/Challenger Model Selection

Speed up testing and validation with autoML that finds the best combination of features and models and automatic selection of the best machine learning model to fit your dataset.

Model Ensembling

Multiple levels of both fully automatic and easily customizable ensembling to increase accuracy and ROI.

Turium Interpretability

AutoML powers a robust Interpretability toolkit to include explanations, visualisations and customizations.

Automatic Label Assignment

Reduce error rates and save time with automatic labelling that predicts the class for every scored record, in addition to returning the per-class probabilities.

Model Validation

Assess model robustness and mitigate risks in production by obtaining a holistic view of the models and preventing failures on new data.

Unsupervised AutoML

Immediately get new insights on your unlabeled data with unsupervised techniques such as clustering to automatically group topics, outlier detection to identify irregularities in your data, and dimensionality reduction to reduce model overfitting and complexity.

Imbalanced Dataset Handling

Improve the accuracy in imbalanced use cases with access to special, proprietary algorithms which emphasise accuracy of rare classes over the more frequent but less valuable classes.

04/

Turium Interpretable AI

Easily understand the ‘why’ behind model predictions to build better models and provide explanations of model output at a global level (across a set of predictions) or at a local level (for an individual prediction).

05/

Generalised Linear Models (GLM)

GLMs are an extension of traditional linear models. They are highly explainable models with the flexibility of the model structure unifying the typical regression methods (such as linear regression and logistic regression for binary classification).

Generalised Additive Models (GAM)

GAM is a Generalised Linear Model (GLM) in which the linear predictor depends on predictor variables and smooth functions of predictor variables.

Generalised Additive Models with two-way interaction terms (GA2M)

GA2M is an extension of GAM which selects the most important interactions between features and includes functions of those pairs of features in the model.

Partial Dependence Plot (PDP)

Plot that shows how a column affects predictions at a global level, with the ability for users to explore columns and how they affect predictions.

Explainable neural networks (XNN)

These neural networks consist of numerous subnetworks, each of which learns an interpretable function of the original features.

Feature Importance

Calculate which features are important for the model’s decision making, both naive and with transformed features.

Skopes Rules

This algorithm learns a simple set of rules for performing classification.

Surrogate Decision Trees

Identify the driving factors of a complex model’s predictions in a very simple, visual and straightforward way.

Individual Conditional Expectation (ICE)

Plot that shows how a column affects predictions at an individual level, with the ability to drill down to any row of choice and compare/contrast with average partial dependence.

Shapley Reason Codes

Provide model explainability at a record level for non-linear models for global and individual records.

Leave One Covariate Out (LOCO)

Identify features that are important to the Surrogate Random Forest predictions from an aggregated or row level view.

k-LIME reason codes

Generate novel reason codes at a record level, subsets of the dataset or at an aggregated level for the entire dataset.

06/

ALGOREUS Act

SECURITY

Role-, Classification-, and Purpose-based paradigms

Integration with existing authorization models

Propagation by default; extreme configurability

DATA HEALTH MONITORING

Pre-built checks, and customizable checks

Leverages Fabriq’s lineage system, for alerting and impact analysis

Full triage & tracking through integration with Fabriq Issues

LINEAGE

Interwoven with security paradigm; provides immutable tracking

Allows for impact analysis, granular usage analysis

Rich APIs allow for navigation upstream and downstream, for a given resource

APIs / EXTENSIBILITY

Custom webhooks and writeback procedures can be authored directly in Turium’s applications

Turium Third Party Authorization framework allows external client applications to be registered with the platform, and fully leverage granular security paradigm

SCENARIOS & SIMULATIONS

Treating Your Business Like Code; branch, simulate, and explore at full scale

Simulations can leverage all types of models and can be tactical or long-lived, refreshing along with new data and models

Turium’s out-of-the-box application for graph/relational exploration of the ontology

Allows for easy creation of new scenarios, and simulate “what-if” conditions

SYNCHRONIZING DECISIONS BACK TO EXTERNAL SYSTEMS

Data egress leverages all of the capabilities of Fabriq’s Data Connection framework

Full lineage is maintained from data to decision, allowing the organisation to always ask “what was the state of the world?” when a particular piece of data or metadata was written externally

04/

Model Repository

05/

Model Management

Create a central place to host and manage all experiments and its associated artefacts, across the entire organisation. Register experiments as models, including both auto generated and custom metadata to have a centralised view of all models.

Model Versioning

Register experiments as new model versions and maintain a transparent view of all deployed versions.

3rd Party Model Support

Manage models trained on any 3rd party framework, including scikit-learn, PyTorch, TensorFlow, XGBoost, LightGBM and more, just like your native models.

04/

Model Deployment

05/

Target Deployments

Build once and deploy to any environment.

Deployment Modes

Deploy models within the production environment in different modes, including multivariant (A/B), champion/challenger and canary. Can be deployed in real-time (hosted RESTful endpoint), in batch (supported source and target datastores), asynchronously or as streaming data.

04/

Model Monitoring

05/

Data and Concept Drift

Maintain model oversight and know if your models are scoring on data they were not meant to or trained on.