Building websites powered by artificial intelligence

We make computers think like humans

Our Story

Imagine if your computer could predict the stock market or tell you which football team will win the game this weekend.  What if it could look at x-rays and detect cancer, or predict crime, forecast hurricanes, and anticipate shark attacks? Up until now, the world of AI (artificial intelligence) has been limited to mathematicians, computer programmers, and engineers. There has never been an easy way for somebody who is not an AI expert to do all of this. 

We spent years working with a programmer creating various neural networks, and every time it would take months of trial and error to make something that sort of worked, but did not seem ideal.  What we did not realize at the time was that there are hundreds of different types of neural networks and algorithms, each having hundreds of possible variations and settings, so even when our programmer was done with the project,  he had really just tested a very small fraction of the possible solutions. It would have taken him many more months to see if there were better ways of doing it, and there was still no good way to know when to stop trying.

In 2016, we realized there must be a better way to do all of this.  Why settle for what might be the 25th best solution, or the 500th best solution, when we could test all the solutions at once? If, for example,  you are trying to predict if Apple Computer stock is going to go up or down, it is easy to test thousands of different AI strategies on real data to see which one actually makes the most money.

So, we created a simple system where we just upload the data (stock prices, photos, medical records, etc.) and our server automatically does the rest. To make the prediction,  it uses neural networks and similar leading-edge technologies. Neural networks are a type of machine learning, where the computer is programmed to "think" like the human brain does. It learns like people do, and because we have access to virtually unlimited processing power via cloud computing, we can train the neural network to be "smarter" than a human. It is like we are able to give it a lifetime's worth of expert experience in a matter of days.

Some of the machine learning algorithms we use include:
k-Nearest Neighbor

Naive Bayes
Classification and Regression Trees
Decision Trees and Random Forests
Discriminant Analysis
Ensemble Techniques (bagging, boosting, etc.)
Bucket of models (Cross-Validation Selection, Gating)
Stacking
Levenberg-marquardt back-propagation
Gradient descent
Gradient descent with momentum
Gradient descent with momentum and adaptive rule back-propagation
Resilient back propagation
Scaled conjugate gradient back-propagation
BFGS quasi-Newton back-propagation
One step secant method
Nonlinear Single Shot Learning Algorithm (NSLA)
Linear quantum Single Shot learning Algorithm (LSSA)
Negative Correlation Learning (NCL)
Design of Experiments (DOE)
Taguchi method
Fuzzy membership functions
Q-Learning and Recurrent Reinforcement Learning
Levenberg-Marquardt BP algorithm
Directed Artificial Bee Colony Algorithm
ANFIS networks with Quantum-behaved Particle Swarm Optimization)
Levenberg–Marquardt (LM) algorithm
Hybrid of fuzzy clustering and TSK fuzzy system
TSK fuzzy system tuned by Simulated annealing improved bacterial chemotaxis optimization
Niche Genetic Algorithm (NGA)
United immune programming (IP)
Gene expression programming (GEP)
Bacterial colony radial basis function neural network (RBFNN)
Bacterial foraging optimization (BFQ)
Tabu search (TS)
Simulated annealing
Geodesic Flow Kernel (GFK)
Subspace Alignment Domain Adaptation (SADA)
Subspace Interpolation Dictionary Learning (SIDL)

We also test different types of neural networks such as:
Radial basis function (RBF) network
Feedforward neural network
Generalized regression neural network
Generic Algorithms (such as NEAT, MuliNEAT, HyperNEAT, and Novelty Search)
Restricted Boltzmann machine (RBM) and also convRBM Convolutional RBM
Principal component analysis (PCA)
Self-Organizing Maps
GSN (goal seeking neuron)
Fuzzy neural network (FNN)
Wavelet Neural Network
Weightless Neural Network
Recursive neural networks
Hopfield network
Bidirectional associative memory (BAM)
Elman SRN
Jordan networks
Echo state network
Liquid state machines
Neural history compressor
Long short term memory (LSTM) network
Bi-directional RNN
Continuous-time RNN
Recurrent Multi-Layer Perceptron (RMLP)
Second order RNN
Multiple timescales recurrent neural network (MTRNN) model
Pollack's sequential cascaded networks
Neural Turing Machines (NTMs)
Neural network pushdown automata (NNPDAs)
Bidirectional associative memory using Markov stepping
Probabilistic neural network (PNN)
Echo state network (ESN)
Elman Network (EN)
Higher Order Neural Network (HONN)
SETAR
Holographic associative memory
Instantaneously trained networks
Spiking neural networks
Dynamic neural networks
Cascading neural networks
Neuro-fuzzy networks
Compositional pattern-producing networks
One-shot associative memory
Hierarchical temporal memory
Markov chains
Boltzmann machines
Autoencoders
Sparse autoencoders
Variational autoencoders
Denoising autoencoders
Deep belief networks
Deep convolutional neural networks
Deconvolutional networks
Deep convolutional inverse graphics networks
Generative adversarial networks
Gated recurrent units
Neural Turing machines
Bidirectional recurrent neural networks
Bidirectional long / short term memory networks
Bidirectional gated recurrent units
Deep residual networks
Extreme learning machines
Liquid state machines
Support vector machines (SVM)
Matching Nets (for one-shot learning)
MANN (for one-shot learning)
Siamese Neural Networks (for one-shot learning)