February 14, 2021

Outsmarting ML Biases: A Checklist

Machine learning algorithms relentlessly search for a solution. In the case of GANs, the generator and discriminator network somehow finds a way to fool each other. The result is a Deepfake. Not that deep fakes are harmless but ML is used in more critical industries such as healthcare. So when a model that is fed with an underrepresented dataset is used, the chances of misdiagnosis increases. “Each ML algorithm has a strategy to answer optimally to your question,” warned Luca. ... The different definitions makes things even more cumbersome for the data scientist. Citing the work on the impossibility of fairness, Luca also explained why some notions of fairness are mutually incompatible and cannot be satisfied simultaneously. “ There is no single universal metric for quantifying fairness that can be applied to all ML problems,” he added. No matter how fool proof the data curation process is, loopholes might creep in. So, what are these loopholes? ... When it comes to ML fairness toolkits, Google’s TensorFlow team has been on the top. The team has been developing multiple tools to assist niche areas within the realms of fairness debate. The whole debate around ML fairness is forcing companies like Google to establish an ecosystem of fairer ML practice through their tools. 

Visual Studio Code comes to Raspberry Pi

There are already some great editors, but nothing of the calibre of VS Code. I can take my $35 computer, plug it into a keyboard and mouse, connect a monitor and a TV and code in a wide range of languages from the same place. I see kids learning Python at school using one tool, then learning web development in an after-school coding club with a different tool. They can now do both in the same application, reducing the cognitive load – they only have to learn one tool, one debugger, one setup. Combine this with the new Raspberry Pi 400 and you have an all-in-one solution to learning to code, reminiscent of my ZX Spectrum of decades ago, but so much more powerful. The second reason is to me the most important — it allows kids to share the same development environment as their grown-ups. Imagine the joy of a 10-year-old coding Python using VS Code on their Raspberry Pi plugged into the family TV, then seeing their Mum working from home coding Python in exactly the same tool on her work laptop as part of her job as an AI engineer or data scientist. It also makes it easier when Mum has to inevitably help with unblocking the issues that always come up with learners.

This new open source tool could improve data quality within the enterprise

While Soda SQL is more geared toward data engineers, Soda also offers a hosted service geared toward the business user and, specifically, the chief data officer (CDO). Interest in data testing and monitoring might start with the CDO when they recognize the need to ensure quality data feeding executive dashboards, machine learning models, and more. At the same time, data engineers, responsible for building data pipelines (transforming, extracting, and preparing data for usage), just need to do some minimal checks to ensure they're not shipping faulty data. Or, you might have a data platform engineer who just wants hands-off monitoring after connecting to the data platform warehouse. In this universe, data testing and data monitoring are two distinct things. In both cases, Baeyens said, "The large majority of people with which we speak have an uncomfortable feeling that they should be doing more with data validation, data testing, and monitoring, but they don't know where to start, or it's just kind of blurry for them." Soda is trying to democratize data monitoring, in particular, by making it easy for non-technical, business-oriented people to build the data monitors.

Cybersecurity is still the #1 risk for manufacturers

We see lots of incidents, but there’s no obligation for the owners and operators to disclose the incident. The incidents that you see in the media are often just a small percentage of the incidents that you actually see in the public eye. We know of many serious incidents that you’ll never read in the headlines and for good reason, really. So, what I would do is say that cybersecurity is still a priority for many organizations. It’s their number one risk, and it’s something that they’re dealing with every day. ... Ask the question, “What is the problem that I’d like to solve, as a result of implementing digital where any other solution couldn’t?” If you’re already on that journey, I would be looking back and reviewing and saying, “Does my digital solution so far answer the question? Is it solving the problem that I want to solve as a result of a digital solution?” In a recent study, we found that less than 20% of organizations have more than a third of the employees actually trained in digital, and trained in their digital strategy as an organization. But, more than 60% of our customers actually have a digital strategy, so there’s a mismatch between customers in heading out on the digital journey, but not really taking their employees with them.

Keeping control of data in the digital supply chain

While organisations will never have as much control over a supplier’s security as they do their own, they can take steps to minimise risks. Security standards must be set out within service level agreements (SLAs), for instance, insisting that the third-party meets ISO 27001 accreditation as a minimum and ensuring that the supplier has a framework of policies and procedures governing information risk management processes. Unfortunately, this approach is rare. The UK Government’s Data Breaches Survey 2019 indicates that less than one in five businesses (18%) demanded that their suppliers have any form of cybersecurity standard or good practice guidelines in place. The issue also becomes more complicated when the sheer scale and intricacy of the average supply chain network comes into play. A firm may have its data stolen from a company three or four connections deep into the supply chain. If the breached third-party lacks the ability to detect an attack itself, a company’s data could be in the hands of criminals for months before they are finally alerted to the breach. Even if a security breach originates with a third party, it will carry just as much of a financial and reputational cost as a direct attack on the organisation’s own network.

Metaethics, Meta-Intelligence And The Rise Of AI

The notion of ethics has evolved. Decisions around right and wrong always depended on human cognition and were guided by popular sentiments and socially acceptable norms. Now, with the rise of AI, machines are slowly taking over human cognition functions, a phenomenon that author Ray Kurzweil predicts will increase over time and culminate in the advent of singularity where machines irrevocably take over humans, possibly at some distant point in the future. This trend is causing technologists, researchers, policymakers and society at large to rethink how we interpret and implement ethics in the age of AI. ... To face the challenges of the future, we also need to develop a new discipline of meta-intelligence by taking inspiration from the concepts of metadata and metaethics. Doing so will help us improve the traceability and trustworthiness of AI-driven insights. The concept of meta-intelligence has been doing the rounds of thought leadership for the last few years, especially led by people thinking about and working on singularity. The pace of technological evolution and the rise of AI has become essential for human progress today. Businesses around the world are getting impacted by the transformative power of these technologies.

Read more here ...