Explainable Artificial Intelligence
&
Structured Machine Learning
Our integrated platform targets all common aspects of machine learning from ETL (Extract Transform Load: Statistical Type Derivation & Meta Data Generation) to EDA (Exploratory Data Analysis). It handles time series modelling, static data analysis, forecasting, dependency analysis & attribution as well as model scoring. The Platform is backed by a dedicated database for integrated geospatial (GIS) and time series data.
Our Machine Learning platform can answer the question: How likely was the past? This avoids backtest overfitting in econometrics and facilitates robust strategy curation without exhausting valuable backtest data. Data Generation Process (DGP) driven development creates more true positive investment strategies in econometrics!
Many platforms focus on either tabular data in factor models (whilst fudging time, lag 1, 2, 3) or they model everything as a time series – including time invariant data. Our platform integrates time series with geospatial data as well as other data sources.
Cross-sectional models facilitate whole of economy modeling using sparse and asynchronous datasets with no explicit quantisation or imputation.
In each human era, explanations of the mind and every cognitive theory are based on a metaphor that reflects the dominant technology at the time. Sigmund Freud chose the steam engine as the metaphor for his model of psychoanalysis. Most people today would probably choose the computer. No matter the metaphor, this approach is always an oversimplification. At Lambda Faktorie we hold that every theory which seeks to reduce intelligence down to a single principle is destined to fail. Language, for instance, is complex. Forming the past tense of a verb requires the application of a system of rules for regular verbs and a system of association for irregular verbs. One simple task: two distinct mechanisms. Knowledge, another example, is an accumulation of facts. Humans can deftly reason about facts, and they can form abstractions from these facts. General intelligence is inherently structured rather than guided by a single principle, if only because the mind comprises over 150 distinct functional areas. Marvin Minsky calls this the “Society of Mind” with different actors responsible for different tasks. Much of Deep Learning today is centred upon end-to-end models with a single homogenous mechanism – in a Freudian sort of way. Therefore Deep Learning struggles, for instance, with inference and abstract reasoning — and often requires inordinate amounts of data for model training. With Deep Learning, knowledge is never encoded, only approximated through “universal function approximation” along with all other input vectors. When a homogenous mechanism applies to a universe of data, mathematical interpretation is lost, yielding black-box systems. The best A.I. researchers solve complex problems through the structured composition of hybrid systems. Alpha Go required a combination of deep learning, reinforcement learning, game theory, and Monte Carlo search – to win. We call this Structured Machine Learning. No single principle! Instead, we pursue the efficient, structured composition of finite hybrid systems: “Infinite Use of Finite Means.”
Predict probable mineral deposits in Google Earth using satellite data, geological data and borehole samples. The platform generates a file with predictions as Keyhole Markup Language (KML) which may simply be imported into Google Earth for display. KML predictions are ranked according to the probability that deposits may be found in a particular location.
Our “Autonomous Supply Chain Manufacturing” project seeks to create a robotic manufacturing network. It composes advanced manufacturing capabilities through artificial intelligence and interfaces these capabilities through an interconnectivity model that is mathematically designed to scale. This enables rapid, independent local manufacture and allows end-users to design bespoke products that are delivered on demand by a dynamic supply chain. United States Patent 10152760: METHODS FOR AN AUTONOMOUS ROBOTIC MANUFACTURING NETWORK and United States Patent 10242316: ROBOTIC CAPABILITY MODEL FOR ARTIFICIAL INTELLIGENCE ASSISTED MANUFACTURING SUPPLY CHAIN PLANNING.
Antarctica is a harbinger of global climate change. Modeling the weather in Antarctica and identifying exogenous drivers of climate in the Antarctic unlocks mysteries of future climate change and helps identify temperature cycles. Shown to the left is a Cross-Sectional, Bayesian Latent Profile Analysis (LPA).
Multiphysics Simulation, Optimization, and Data Analytics. “Hypersonics” comprises aircraft and “glide vehicles” traveling at over 5 times the speed of sound (Mach 5). Specifically, the hypersonic regime is defined as the realm of speed wherein the physics of flows is dominated by aerodynamic heating. This poses several engineering problems, including adaptive flight control techniques. Our approach to modeling hypersonic vehicle flight is to combine computational fluid dynamics & multiphysics simulations with modern machine learning techniques.
Directly Left: Lockheed Martin SR-72 (United States )
Bottom Left: Avangard Hypersonic Glide Vehicle (Russia)
Bottom Right: Lippisch P13a Ramjet (Germany, WWII)
Finding hidden patterns in complex data using information theoretical criteria and SEM (Structural Equation Modelling).
How many clusters are really in the data?