Data Science Insider: March 25th, 2022

In This Week’s SuperDataScience Newsletter: Nvidia to Introduce New Enterprise AI System. Misuse of AI datasets may cause biases and compromise the algorithms. How the Immune System Model Can Enable Deep Neutral Networks To Thwart Attacks. The Machine Learning Program that Will Enhance Speech Recognition. AI System that Can Model Chemical Weapons In Just a Few Hours. Cheers, - The SuperDataScience Team P.S. Have friends and colleagues who could benefit from these weekly updates? Send them to [this link]( to subscribe to the Data Science Insider. --------------------------------------------------------------- [Nvidia’s New Enterprise AI System]( brief: Partnering with prominent virtual infrastructure providers, such as VMware, Red Hat, and Domino Data Lab, Nvidia has launched Nvidia AI Enterprise 2.0, a version of the vendor's cloud-native suite that is about to transform the existing approaches towards cloud systems and data centers. Enterprise 2.0 has the capacity of supporting data servers, systems, and online infrastructures of varying complexity. Additionally, as the Nvidia team introduced the new GPU products to leverage the complexity of the AI stack, they also managed to devise an integration plan to help IT professionals transition from working with CPU-only cluster architecture to GPU technologies. This is a significant step as GPU stack is more advanced, which is why Nvidia worked hard to implement a smoother transition for the data scientists and other users. "It's about enabling the IT teams to leverage their existing skill sets and apply and leverage a new technology domain, like the GPUs," Chirag Dekate, an analyst at Gartner said. Why this is important: The data science industry is very fast-paced and is constantly evolving. For any data science professional, it's crucial to stay up to date with new technologies, updates, and approaches. At the same time, this new product is a key step towards more advanced AI databases. This initiative aims to aid companies and enterprises in improving their existing processes and systems, and to help their IT teams use their existing transferrable skills to transition to GPU technologies. [Click here to sign up!]( [The Dangers of Misusing AI Datasets]( brief: Known as the "off-label" use, this article discusses the practice of using data from one study to replicate and train AI for a different study, which may not always reflect the algorithms accurately. Even though there is a common perception that databases are being fed with raw images only, this is often not the case. Developers can also download existing datasets with MR images that have already been processed, synthesize raw measurements from those sets, and develop algorithms based on those synthesized images instead of inputting actual raw data. "Our results showed that all the algorithms behave similarly: When implemented to processed data, they generate images that seem to look good, but they appear different from the original, non-processed images," said study lead author Efrat Shimron, a postdoctoral researcher in Lustig's lab. "The difference is highly correlated with the extent of data processing." Why this is important: Using this method can lead to a significant bias in the findings as the images generated by the MRI reconstruction algorithm from the fastMRI dataset look better than the originals and appear to be 48% clearer, which in turn does not warrant an adequate representation of real data. Therefore, it becomes more challenging to obtain reliable data, diminishing the public confidence in the accuracy of the AI learning processes. [Click here to read on!]( [Enabling Deep Neutral Networks to Thwart Attacks]( In brief: The term "deep neural network" refers to a set of machine learning algorithms such as image classification, language processing, vision, fraud detection, etc. Using the human immune system model as an inspiration, researchers at the University of Michigan applied the operations and processes of the body's defences against pathogens to inform the creation of defenses for neural networks. Specifically, they chose to model the process of B-cells fighting antigens in the tissues of genetically modified mice, turning it into code to test the Robust Adversarial Immune-Inspired Learning System (RAILS) defences that were triggered in response to adversarial data inputs. When the results were compared, the RAILS defence-based algorithm was very similar to the immune system process, but most importantly, it outperformed two other common machine learning processes used to combat attacks, known as Robust Deep k-Nearest Neighbor and convolutional neural networks. In addition, the overall accuracy of RAILS has improved significantly thanks to the use of the immune-inspired defense system model. Why this is important: This research shows the knowledge of human biological processes can inform machine learning, which may suggest a strong correlation and more possibilities for further examination of how the models of biological processes can inform AI. Moreover, this process of discovery can go both ways: as researchers continue to make a lot of progress with machine learning, it potentially can give a better understanding of the human condition. [Click here to discover more!]( [Machine Learning Enhances Speech Recognition]( In brief: Hearing aids today are programmed with codes that merely detect sound waves, but the current algorithms still ignore the complexities of modern-day speech. Researchers Roßbach, Meyer, and others decided to tackle this problem by developing a machine-learning algorithm that determines how the listener can perceive sounds in any given environment, depending on the acoustic conditions they experience in the moment. The model uses an automated speech recognition system that has been trained and tested with the use of speech recordings in different noisy environments, trying to capture how individuals with hearing impairments perceive it. The recordings were played to provide data for the machine learning system; those recordings were also played for normal-hearing and hearing-impaired listeners to test how they perceive them compared to the machine, resulting in a 50% word-error rate for each listener, regardless of their capabilities. Why this is important: Even though hearing aids had been an important invention that made life easier for persons with hearing disabilities, they still have many limitations. As much as the new developments represent to the scientific society and the world at large the positive impact of progress, it is important to use these new technologies to address social inequalities that many marginalized groups still experience to this day. [Click here to see the full picture!]( [AI can Model Chemical Weapons in Hours]( In brief: The research team from Collaborations Pharmaceuticals, In., King's College London, and Collaborations Pharmaceuticals discovered that AI systems currently operated by pharmaceutical companies daily to develop new drugs can easily become subject to misuse. To prove that, the group examined the system's filters used to exclude toxic drugs from the search, tweaking them in reverse to find these types of drugs specifically. At the end of the search that ran for six hours, the AI system MegaSyn had displayed 40,000 results featuring toxic drugs that could also be turned into potentially fatal chemical weapons. The researchers specifically attempted to search for drugs related to VX, a toxic chemical that attacks the muscles involved in breathing. The AI software not only had found drugs similar to VX but also discovered some agents that were even more toxic than VX. Why this is important: Understanding vulnerabilities of current AI systems, especially those used by mass pharmaceutical retailers, is essential in terms of public health safety and overall security. As the researchers themselves stated, this discovery is eye-opening, especially for anyone working in the fields related to the AI and drug industries. It is important to ensure that all measures are taken to prevent any potential abuses of this type of data that can be used to create dangerous and illegal chemical warfare. [Click here to find out more!]( [Super Data Science podcast]( this week's [Super Data Science Podcast](, GPT-3 expert and PhD student Melanie Subbiah joins us to discuss her recent breakthroughs in natural language processing, which sent shockwaves through the mainstream media back in 2020. --------------------------------------------------------------- What is the Data Science Insider? This email is a briefing of the week's most disruptive, interesting, and useful resources curated by the SuperDataScience team for Data Scientists who want to take their careers to the next level. Want to take your data science skills to the next level? Check out the [SuperDataScience platform]( and sign up for membership today! Know someone who would benefit from getting The Data Science Insider? Send them [this link to sign up.]( # # If you wish to stop receiving our emails or change your subscription options, please [Manage Your Subscription]( SuperDataScience Pty Ltd (ABN 91 617 928 131), 15 Macleay Crescent, Pacific Paradise, QLD 4564, Australia