The view of the Principality of Monaco at sunrise
It has to do with logic, but not everything is straightforward. Building artificial intelligence models requires a strict approach to algorithmic logic based on the determining and determining factors that make up the model itself. But there is a word in that sentence that is very important and is determinant. As we now extend all forms of he (generating, predictive, reactive and so on) to include the notion of agents with their autonomous ability to perform in non-defining environments (where the “defined nature” of the calculation is left more open), we need to embrace a new degree of managed chance.
The deliberate use of random numbers in computers science is nothing new. Even basic (by which we mean basic) Programming languages in use over half a century ago will include algorithms to generate random numbers to use in games, cryptography and programming elements needed. While there are many modern techniques for casual numbers and variable random generation, the agent space and it can now be in perfect time to visit Monte Carlo.
What is a Monte Carlo simulation?
A Monte Carlo experiment is a method used to present coincidence in defining models differently where the probability of different results is difficult to measure without injection of a discreet dose of risk and insecurity. As Kenton explains Investopedia“Monte Carlo’s simulations can be applied to a number of problems in many areas, including investment, business, physics and engineering. He has also referred to as a multiple probability simulation.” Name, perhaps predictable, comes from the city to Monaco, I Who is famous for his casinos and gambling among other things.
Larger financial institutions use Monte Carlo methods as part of models that support tasks such as pension, portfolio and other financial assets. Monte Carlo’s methods are essential for many trade efforts, ranging from banking financial risk modeling to predicting the bringing of advanced materials to air space applications. So how would a case of use play and what lessons teach us about how we use randomization on technology platforms today?
Stock market shakes
Let’s imagine that we want to predict the price of one stock annually from now. We can try to predict how the price of shares will change every day trading due to the random causes of the world and factors such as the daily news cycle, market stories, trade wars, other pandemia, etc.
“We can repeat this process, using price change on a given day as the starting price for the next day of trading; in that way, we can get a stock price path based on the sequence of daily price changes. If we were to repeat this process a million times, with each road to be a sequence of five days, we would end with a million possible price price prices five days in the future, with each The price that represents a different way possible that the market market could get these millions of final stock prices paint a picture of the next price. An invaluable method for modeling phenomena that have a component of destiny, ”explained Phillip Stanley-Marbell, founder and CEO of signoid.
Monte Carlo methods include algorithms recapture repeatedly with slightly different inputs to reach a point where the simulation direction for more repetition would no longer make a change in the general model; This point is often called the convergence point and can take millions of repetitions before it is reached.
For some of the most complex models used in fields ranging from modeling of microprocessor chips, to finance, up to the airspace, convergence achievement can take hours or days, receiving significant computing and energy sources. The main way that organizations accelerate Monte Carlo’s simulations is by doubling the investment of resources and using numerous cores in processors or graphic processing units (GPU).
Stanley-Marbell says there is a way to achieve the same results as a converged Monte Carlo without using the use of brutal force of increasing numbers in the processor and graphic processing units. Its company provides hardware -based computer equipment designed (to enable organizations in financial services and manufacturing industries) to speed up the workloads currently using Monte Carlo methods.
He says processors and GPUs perform calculations on numbers (eg “7” or “0.0094”), but if processors and GPU can perform direct calculations in probability distributions (think of a probability distribution as a value represented With a histogram, perhaps with multiple tops, a long tail, etc.), they will be able to mathematically shorten the repeated and tedious steps associated with Monte Carlo methods, without losing quality and correctness and bypassing the need to repeat until convergence achieved.
Converging in the future, faster
Modern computing solutions have shown that organizations do not only need to rely on Monte Carlo’s simulations to achieve the desired results through convergence. The Signitoid Computing Accelerator is said to allow organizations to achieve the same types of results through a single -transition execution, resulting in faster results without the challenges of determining if a model has reached convergence.
“Distilled at its basic level, this approach includes a paradigm change where instead of re -evaluating numerous simulations to achieve convergence, you calculate all possible paths at the same time to achieve a response, each time. For Many workloads, using a determining solution provides results up to 1000 times faster, cutting out computing time from hours to seconds, allowing financial institutions to apply potentially to Important real-time business operations that they previously needed to run as overnight groups, “detailed Stanley-Marbell.
He suggests that these “dramatic speeds” are achieved through three main developments. The first involves rethinking how the calculations are realized. In addition to performing calculations in numbers such as “7” or “0.0094”, as do traditional processors or GPUs, the signoid accelerator also performs direct calculations in probability distributions.
Second and third major innovations are a group of efficient digital representations pending patent for these probability distributions that are said to be more efficiently scaled in their hardware requirements than Monte Carlo methods when applied In increasingly larger work loads and effective calculation routines to perform calculations for these representations digital.
A one -way execution
“To say all this simply, the status quo is performing Monte Carlo simulations until things stabilize or reach convergence. Transfer to a single -transition execution allows organizations such as protective funds and investment banks to meet their existing critical business loads that today use Monte Carlo methods, with higher loyalty or more sophisticated models, enabling They make better portfolio models, better risk analysis, better interest rate models and ultimately to generate better returns, ”he said Stanley-Marbell.
While the world of algorithmic functions and the rise of the agent it is discussed more widely in the context of the models that move between the determinant and non-determining foundations, it is perhaps useful to have a meaning where Monte Carlo methods adapt to the logic of probability. Even if technologies like accelerators displayed here means that we never go to Monte Carlo, however, knowing that place/practice exists can simply make us not go.