Technology

Technology

Even though Neural-Lotto’s core is based on neural network technology, there is much more behind its phenomenal performance. The NeuralReality AI Engine employs state-of-the-art genetic search algorithms (which mimic the process of natural evolution) that have been specially engineered to avoid converging to local optima. The level of complex problem solving and pattern searching and matching that Neural-Lotto can achieve is unprecedented.

There are up to 999 hidden layers each with up to 990 neurons, coupled with more than 20,000 perceptrons which results in over 1 million active neurons. This complex network features dynamic multithreaded backpropagation, evolutionary algorithmic and gene expression programming, probabilistic metaheuristics and expectation-maximization, with a core non-parametric statistical model.

There are also 4 auxiliary self-organizing associative Kohonen/Hopfield hybrid networks and 2 Bayesian probabilistic networks assisting multiple core functions, with highly evolved distributed representation and asynchronous control mechanisms, content-addressable memory matrixes and a fully fault-tolerant architecture.

These are paired with numerous search & discover heuristics, symbolic reasoning, and statistical reasoning techniques including Best-First Seach, Means-Ends Analysis, Nonmonotonic Reasoning, Depth-First and Breadth-First Search, and 3 Dempster-Shafer implementations, resulting in 5 highly adaptive, fuzzy logic artificial intelligence learning algorithms independently selectable within Neural-Lotto.

Neural-Lotto’s main core is programmed in PROLOG, with many auxiliary systems programmed in LISP. Speed-critical processes are programmed in assembly language, with all other interfaces and subsystems programmed in C++. Relational data storage is managed by Oracle, SQL Server and DB2 DBMSs, including neural network synapses (axons and dendrites), content memory and historical lottery data. MySQL and PHP provide an on-line web user inter­face for remote neural net­work input para­me­ter configuration.

As for hardware, Neural-Lotto runs on a custom-made Quad 6386 SE Piledriver–based 2.8GHz (3.5Ghz max) 16-core x 4 (256 core) supercomputer, with 32 cores (2 processors) devoted to database-related tasks and 224 cores (14 processors) plus 4 NVIDIA Tesla K40 accelerator modules (2880 CUDA cores each) with a staggering 21.87 TeraFLOPs of combined peak power, devoted to neural-related tasks. Each node (4 CPUs, 64 cores, 1 Tesla K40) has access to 1TB of RAM, for a grand total of 4TB of memory devoted exclusively to Neural-Lotto.

Hard drive storage is managed by 40 arrays of 10 enterprise-class Savvio 600GB 15K rpm SAS hard drives, boasting 240TB of mass storage, with an additional array of 10 drives (6TB) providing redundant fault-tolerant backup for critical neural network processes. A 16kW diesel generator provides unattended, uninterrupted backup power.

The NeuralReality AI Engine — the Neural-Lotto core, took nearly 150,000 man-hours to program, test and deploy, and more than US$3.2 million to implement. Over the course of 8 years, Neural-Lotto underwent many software and hardware changes and upgrades, aimed soley at improving overall performance, stability and efficiency.

The result — the world’s most advanced pattern-matching, search & discover artificial intelligence neural network applied to lotteries ever created.  

Neural-Lotto, along with its predecessor NeuralAnalyzer i7000, has proven without a doubt that there is no such thing as blind luck, pure chance or random coincidence — that even a seemlingly random event such as a lottery drawing has within itself an inherently subtle pattern embedded into it common to all matter: nothing in this universe is created by mere chance or accident. Now we are not saying the lottery is rigged. We are saying that seemingly random events such as tossing a dice 100, 1,000 or even 10,000 times yields results with a definite, subtle pattern, if not at all apparent, and that with enough historic data and computational processing power, the next toss or tosses can be predicted to a large degree.