The World’s First Real Neural Computers Offer Actual Intelligence
Expansion of Deep Learning, and Knowledge Enhancement
For Immediate Release: April 4, 2022
Actual, not artificial, intelligence, learning, and knowledge is inherent in all forms of life. It is a standard feature, and natural capability of all nerve cells, also known as neurons. Living neurons can process, store, retrieve, edit, and delete data. They perform these tasks using both external and internally provided data and guidance.
However, totally different than an inanimate object, like a silicon chip, neurons can self-create new intelligence, knowledge, and learning.
It is true that digitally simulated, and neural chip emulated networks can function as if they are neurons, but these digitalized imitations have many limitations. In the end, natural live neurons will always outperform the inanimate computer chip when performing learning, being intelligent, and expanding the realms of knowledge.
Knowing this, mankind has for decades attempted to establish a means of copying these amazing capabilities of nature. That quest, to deliver a 1,000-fold magnitude leap in data processing performance, has now been accomplished.
BCM Industries (BCM) has established a process and developed the hardware and systems to allows mankind to directly utilize and benefit from the management and control of millions of live neurons. The result is organized live neural networks with near blinding processing speed and throughput, massive data storage capacities. and lightening data transfer rates, all controllable with a laptop computer.
Tissue Computing with “Live Neurons” will soon replace
Digital Computing in Most Large Deep Learning Applications
The forthcoming availably of a family of Tissue Operating Device (TOD™) Models, which include Tissuing Computing, using live neurons, will soon replace digital computing in many large deep learning applications.
The availably of TOD™ Tissue Computing is extremely timely because the rapidly expanding need for large scale deep learning processing is rapidly surpassing the ability of any inanimate objects, the silicon chip, to fulfill the demand for processing power.
Current demand is approaching the limits of the physical processing capabilities of the digital, chip-based, computer. The result is either: (a) these expanding deep leaning applications will never occur; or (b) the number of individual parameters processed will need to be reduced, which will affect the accuracy and quality of results.
To address these limitations and others we explore the deep learning use and needs of GPT-3, a neural network machine learning model. This is a popular third version of the Generative Pre-trained Transformer. The transformer deep learning model applies over 175 billion machine learning parameters to each usage.
Although the GPT-3 is a great advancement, it has issues and problems. One big issue is that GPT-3 is not constantly learning. It has been pre-trained, which means that it doesn't have an ongoing long-term memory that learns from each interaction. In addition, GPT-3 lacks the ability to explain and interpret why certain inputs result in specific outputs. Other major deficiencies include the fact that GPT-3 only accepts input text a few sentences in length and that it requires a long time to process the information and generate results.
These are all common issues and concerns in deep learning platforms. A review of the computational demands of deep learning applications in five prominent application areas showed that progress in all five is strongly reliant on increases in computing power.
These are all common issues and concerns in deep learning platforms. A review of the computational demands of deep learning applications in five prominent application areas showed that progress in all five is strongly reliant on increases in computing power.
Extrapolating forward, this reliance reveals that progress in deep learning is rapidly becoming economically, and technically unsustainable. Thus, continued progress will come from dramatically more processing power, or a results-damaging degradation in the computationally-efficient methods.
As illustrated in the Figure, the just-in-time solution to these issues is moving deep learning applications from digital platforms to a live neural processing platform such as provided by a TOD™ Tissue Computer.
A graphical presentation of the approaching digital processing capacity limits and those of the solution, a TOD™ Tissue Computer platform, is illustrated in the Figure.
Experience has demonstrated that deep learning is massively reliant upon computing power. This is due, in part, to the number of parameters involved, and how performance scales as additional training data is added. Deep learning will also be enhanced, with a TOD™ Tissue Computer platform which delivers ongoing long-term memory capabilities.
It has been proven that significant benefits are achieved by the neural network containing more parameters than there are data points. The challenge to adding more parameters is the number of deep learning parameters must grow as the number of data points grows.
This means the computational requirements grow at least by the square of the number of data points. This quadratic scaling, however, is an underestimate, the reality is the growth must be much larger.
Extreme dependence on massive computing power for improved performance is not unique to deep learning. It is also present in all areas of deep learning. Examples are weather prediction, oil exploration, data mining and any processing activities involving Big Data.
TOD™ Tissue Computers naturally address all of the key requirements for handling Big Data – “real” rather than artificial intelligence, built in short and long-term memory, the ability to process sensory inputs (visual, auditory, etc.) rather than requiring these inputs to be digitalized (thus losing vital detail and processing time), and of course the blistering speed and natural power of a real neural computer.
For additional information on Tissue Computing technology, current neuron processing activities, TOD™ design, and related subjects, visit the BCM Industries website or contact BCM.
