How Ai is underpinning the future
on Mar 09, 2020
Many users might be unaware of just how much AI technology already affects and informs their day to day lives. From facilitating interactions with smartphones and smart speakers, to enhancing online shopping and search engines – AI is already here, underpinning processes and making them smarter and faster. And yet, these applications are only the tip of the iceberg, and the time is fast approaching when our current uses for AI will come to look very simplistic indeed.
The processing power, in-memory computing and memory performance requirements that AI demands are all extensive, however, and infrastructure that can handle constant flows of complicated, data-intensive processes must be established before the AI dream can be properly realized. That’s where Samsung’s HBM2E Flashbolt solution comes in. The high-bandwidth memory innovation is effective when it comes to helping AI draw complex patterns from large volumes of data, and the low latency storage devices that are compatible with it – including Samsung’s low latency SSD (Z-SSD) and high bandwidth PCIe Gen4 SSD (PM1733) – are designed to handle AI and high-performance computing (HPC) workloads with ease.
Below, we unpack the enormously transformative trend that is AI, the capabilities the technology will give rise to, and the ways in which – empowered by next-level memory solutions – it will come to inform and empower our lives.
The ability of AI systems to recognize and understand images and speech has advanced significantly in recent years. AI can now recognize specific image contexts, allowing it to do things like guess how a person is feeling just from their facial expression. When it comes to creation, AI has the power to transform visual information into novel, highly realistic images and videos. As far as speech recognition, deep learning methods allow AI systems to recognize hundreds of billions of word combinations and respond accurately to user commands within a matter of seconds.
AI learning has been revolutionized by the ‘reinforcement machine learning’ process. Whereas previous ‘supervised’ machine learning methods required sub-optimal actions to be explicitly correct, reinforcement machine learning simultaneously balances the exploration of new courses of action and exploitation of existing knowledge. This allows the system to measure current behavior against the achievement of stated goals and adjust processes accordingly.
The area of AI reasoning and inference requires systems to understand the contextual meaning of information in order to infer entirely new propositions pertaining to it, with recent advancements in the sector enabling AI to answer ‘human questions’ following analysis and combination of vast amounts of data. Again, Samsung’s HBM2E line has been brought to bear here, providing the tools for AI software to learn how to identify and make sense of the complicated patterns gleaned from large quantities of data.
Human-level planning and decision-making require the decision-maker to factor the long-term ramifications of their decisions into their planning. Research currently being done in the industry will allow contemporary AI solutions to make these kinds of ‘big picture’ decisions, even enabling an AI system to choose a course of action that will prove detrimental in the short term if it believes the behavior will ultimately assist in the achievement of its goal.
AI and Us
Recent increases in computing power and improvements to AI algorithms have made it possible to run machine learning solutions on smartphones instead of in the cloud. The inception of on-device AI not only drastically accelerates processing, but also makes devices more secure and improves data-fetching capabilities in areas with poor connectivity. Some functions that on-device AI will enable include:
• AI-powered Cameras: AI-capable cameras provide superior face-detection capabilities and improve white balance and exposure settings, while the AI’s neural processing unit detects what kind of scene is being captured and optimizes the camera features accordingly.
• extended Reality (XR): In allowing for the creation of an intuitive user interface that responds to speech and movement patterns, helping devices understand the 3D world and allowing for the addition of enhanced interactivity, AI will prove the key to unlocking both virtual and augmented reality on our personal devices.
• Virtual Assistants: On-device AI will prevent virtual assistant applications from having to process data in the cloud, allowing them to perform smarter and faster than ever before.
The potential of high-powered computing (HPC) also stands to be unlocked by the inception of AI. HPC, in turn, will enable things like live streaming services, which require huge quantities of data to be processed in real-time, live weather tracking, product testing and the live analysis of stock trends. HPC is set to save enterprises money as well as the increased sophistication of cluster architecture increases efficiency and lowers TCO (total cost of ownership).
AI will also prove all but a necessity for the next generation of automotive solutions. In the cars of the future, AI will anticipate impending component failures and identify and act upon road hazards to avoid accidents. What’s more, driver monitoring systems will tailor preferences for passengers while monitoring the driver for signs of distraction or drowsiness. Samsung’s Flashbolt solution proves especially optimal for the automotive sector, offering transfer speeds of up to 3.2Gbps and bandwidth of approximately 410GB/s - a 1.3X enhancement over that offered by its predecessor, the Aquabolt.
Another area that stands to be overhauled by the introduction of AI is that of TV and gaming. Amid the introduction of 8K TVs, Samsung will take 4K, 1080p and 720p videos and use AI to upscale them and make them suitable for 8K TVs. What’s more AI-powered televisions will recommend new content based on users’ viewing habits and be controllable through voice commands.
What AI Still Needs
Despite recent advancements, further development will be needed before AI technology can truly be brought to fruition. HPC will need to be further developed in order to enable the high-performance data analytics required to train machine learning models, and parallel processing will need to be leveraged if the advanced, large-scale applications that will allow AI to evolve are to be realized.
‘Big data’ advancements are another prerequisite for AI development, as they will allow machine learning algorithms to extract patterns from data. Other important requirements for the growth of AI include the development of more sophisticated artificial neural networks and the aforementioned on-device AI capabilities, which will be important in allowing systems to learn.
Underpinned by Samsung Memory
Just as AI will empower the incredible innovations of the future, Samsung’s cutting-edge memory will provide the foundation that AI needs to grow. In January of 2020, Samsung launched its HBM2E Flashbolt – the industry’s first third-generation, 16GB solution of its kind for high-performance computing. In addition to being optimal for HPC systems, the solution will underpin supercomputer advancements, enable AI-driven data analytics and empower state-of-the-art graphics systems. The Flashbolt can reliably offer superior data transfer speeds of approximately 3.2Gbps and memory bandwidth of up to 410 GB/s.
Complementing Samsung’s high bandwidth memory solutions are its low-latency storage devices, including the Z-SSD and PM1733 offerings. The Z-SSD is optimized for AI workloads and HPC and combines extremely low latency with exceptional performance and reliability in one simple-to-use device. Samsung’s Gen4 PM1733 SSD, meanwhile, boasts double the speed of Gen3 SSDs, offering a read speed of up to 8GB/s and write speeds as high as 3.8GB/s, making it optimal for both server and storage applications.