Video instructions and help with filling out and completing Will Form 2220 Compute

Instructions and Help about Will Form 2220 Compute

Hi thanks for tuning into singularity prosperity this video is the 11th in a multi-part series discussing computing in this video we'll be discussing what cognitive computing is current cognitive computing initiatives and the impact they will have on the field of computing Music the human brain is truly an amazing machine able to operate in parallel malleable and fault-tolerant having 100 billion neurons with each neuron having 100 to 1000 synapses synapses being the connections to other neurons this equates to 100 trillion up to 1 quadrillion synapses all only requiring 20 watts of power and the space of 2 liters as discussed in a previous video in this series about computing performance the human brain is postulated to equate to 1 exaflop the performance and other words 1 billion calculations per second and there are many initiatives to reach this excess skill performance by 2020 and supercomputers around the world for us to simulate the brain that being every neuron and synapse in the brain with these exascale systems will require approves a 1.5 million processors and over 1.6 petabytes of main high-speed memory using power in the order of megawatts per hour and taking up the space of entire buildings all of this as compared to our brains that are acquire just 20 watts of power in the space of 2 litres and will still outperform these machines that orders a magnitude faster on the petaflop k supercomputer in japan running neural simulation technology nest algorithms requires roughly three point six eight years to simulate one day a brain activity that's 1,700 times slower than the brain japan's post k exaflop supercomputer aims to increases to 310 times slower simulating one day in the brain and 310 days while these simulations will aid us in unlocking secrets of the brain due to the vast architecture differences between modern computers and biological brains these exascale systems will still be limited in functionality every computer in the world today is based upon volume and architecture having computation and memory fairly isolated with a data bus connecting them whereas biological systems have memory and processing tightly coupled together while von Neumann architecture is still the best choice for the majority of computing applications as seen by the drastic performance differences and brain simulations a more biologically representative architecture has to be implemented neuromorphic architecture first and foremost neuromorphic architectures will allow us to accurately and in real time simulate aspects of the brain however while this is one goal of this new brain inspired architecture our brains are the perfect in any regard they get bored distracted fatigued are biased and are not perfect decision makers they can be inconsistent and prone to errors this then leads us to another goal of neuromorphic architectures to be paired with our devices and accelerate the field of artificial intelligence that is to take the best aspects of the brains functionality and pair them with current computing volume and architecture this all is encompassed under heterogeneous architecture which we discussed in a previous video in this series more multiple compute devices and architectures work in unison together let's look at this in terms of the two halves of the brain the left and right brain the left brain is focused on logic encompassing analytical thinking language and other such tasks while the right brain is focused on creativity encompassing pattern recognition learning reasoning and so on the right brain is clearly more abstract than its left brain counterparts ik waiting to computing left brain tasks are best suited to be handled by traditional computers or the right brain is what neuromorphic computing aims to handle the left brain performance is flops driven while the right brain is driven by converting senses to action or what some call stops synaptic operations per second under HSA heterogeneous architecture the melding of these two halves n is what will lead to truly intelligent robotics and machines that are able to operate in real-time computing devices based on your morphic architectures will be able to truly learn and reason from their inputs especially when paired with optimized software algorithms this has been the epitome of our discussions in this computing series hardware and software tightly coupled together to yield massive performance and efficiency games one such field of computer science that has gained tremendous steam in the past decade and as a basis of how our brains operate as machine learning by creating nodes essentially neurons assigning weights to them and then feeding in large sets of data these nodes begin to interconnect amongst each other like synapses connecting neurons and to vast neural Nets these neural nuts are referred to as machine learning models which can then be applied to our devices and also continually adapt by processing more data this was an extremely quick overview on machine learning and a much more in-depth discussion will be had in this channels a eyes coming back to heterogeneous architecture while neuromorphic chips paired with machine learning models will be able to learn and reason on the von Neumann architecture side these traditional compute devices as we all know excel at repetitive tasks so in this case executing the models produced by the neuromorphic chips neuromorphic chips paired with the traditional computing technologies are leading to a new era of computing cognitive computing the first step in a long road and emulating consciousness and machines Music so are we to design hardware that resembles the human brain well first let's take a brief neuroscience lesson the basics of the composition of a neuron are the cell body axon and synapses translating to hardware terms the cell body is the processor axons are a data bus and synapses are the memory with all three composed to form a nurse and after core essentially a neurosynaptic cores are the nodes and machine learning neural nets but represented through physical hardware rather than software abstraction this alone will present a

Loading, please wait...