Artificial Neural Networks (ANN), brain-inspired simulations, have finally reached their paramount position in the scientific world. Their applications abound as their architecture grows more sophisticated and becomes suitable for new purposes. See our overview of the latest ANN trends.
A human follows the nature in search for new forms and structures. Architects worshiping its smooth lines gave the world Sagrada Familia and Lotus Temple along with Atomium and Helix Bridge. Most definitely, the influence of natural patterns goes far beyond civil architecture. Velcro (hook-and-loop fastener) saw the light after its creator, George de Mestral, studied thistle under a microscope. The idea of drug-reducing coating for ships came to the NASA scientists while observing the sharkskin. Actually, the coating proved to be so efficient, that the technology was banned in sailing competitions.
Computer science long stayed far from the organic world. However, the breakthroughs in neurology and engineering of the 1950’s contributed to a new vision of machine-nature relations. Functional resemblance with a human brain was enough for the scientists to launch a chain of meaningful deployments: from Rosenblatt’s Perceptron through Fukushima’s Cognitron and beyond, to the today’s amazing deep learning breakthroughs. Top-notch applications serve equally effective when implemented either for mundane tasks or for sophisticated analytics.
The industry already receives strong public attention given the driver shortage problem alongside the overall labor shortages in Europe. Numerous recent reports indicate the progress, e.g. NVIDIA reveals its success in application of Generative Adversarial Networks (GANs) for better machine learning. Due to NVIDIA’s recent discoveries, better and faster image segmentation by the AI has been achieved to ensure safer autonomous driving and more accurate road mapping. One more piece of good news is that autonomous semi-truck by TuSimple has recently shown smooth operation in heavy downpour. A light in the tunnel has reached us, finally, after many controversial debates on the autonomous driving in bad weather.
Speech and character recognition
Scientists do their best to replicate human communication patterns in human-machine relations. By now, their endeavors let us speak with our devices, which is more natural for humans than text messaging. We have Siri, Alexa and Google Assistant, ready at any time to react to our voice commands. Nevertheless, speech recognition technologies grow out of their initial and rather limited application. Today’s voice assistants help healthcare institutions keep records, thus unleashing tons of time for face-to-face communication between medical personnel and patients. This is a close to revolutionary improvement, as recent report by Annals of Family Medicine shows that primary care physicians (PCP) spend more than half of their workday on electronic health records (EHR) or other administrative tasks.
ANN superiority over conventional computing methods may yet be questioned. However, ‘artificial brain’ solves a number of industrial challenges much better than its digital predecessor. The ANN as a strong forecasting tool allows for failure prediction to reduce costs, increase efficiency and drive revenues.
The ANN-based failure prediction system by Qulix Systems was introduced specifically for the companies of the railway transportation industry. When the engine shuts down prior to departure or during the trip, the company faces extensive reimbursements, pricy engine repair, and goodwill damages to make it even worse. Qulix Systems offers a solution consisting of a set of sensors, a cloud storage and an ANN. The system analyzes the huge volume of the data on engine behavior and finds the anomalies, if any. Such measures allow predicting engine’s failure before it actually occurs and launching respective maintenance procedures.
So, what the future holds?
The ability to learn by example makes the ANNs a very powerful analytical tool. Its application range is almost unlimited and is constantly expanding. The ANN-based system doesn’t require reprogramming for a special task, although various architectures match better their distinct purposes. Backpropagation and Neocognitron show their best when applied for character recognition, while facial recognition is performed better using Principal Component Analysis. One-size-fits-all solution is unlikely to be discovered, which adds up to the list of other pending issues, both of functional and of ethical nature.
- Will the facial detection of criminals be 100% accurate?
A report published by Big Brother Watch states that the results of the facial recognition CCTV deployment are yet far from acceptable: the technology had picked the wrong person nine times out of 10 as it was tested during public events.
- Will advances in technology endanger our privacy?
A ubiquitous launch of smart CCTV means officials will have access to tons of private data. No system can be absolutely hacker-attack resistant.
- Are we able to create a truly human-like ANN?
As Facebook’s head of AI, Yann Lecun, put it, today we hardly can compare robots even to rats in terms of unfocused intelligence. “Even so, please note that rats are actually quite smart”, Terence Mills comments on the issue in his Forbes article.
These and others questions require non-biased and balanced decisions. However, the ANNs’ beneficial effect for multiple industries is unquestionable and, what’s more it is measurable. Despite many controversies, this fact alone ensures its bright business future.