1. Organizations that want to get started quickly with machine learning may be interested in investigating emerging low-code options for AI.
2. While low-code techniques will never completely replace hand-coded systems, they can help accelerate smaller, less experienced data science teams, as well as help with prototyping for professional data scientists.
LOW CODE
1. First of all, what is low-code? Well, the phrase can mean different things to different people, and its applicability to AI is not entirely nailed down.
2. Mainstream developers have been using low-code (or no-code) approaches to creating business and consumer applications for years, and that largely forms the basis for low-code approaches in AI.
3. Twenty years ago, fourth generation languages (4GLs) were popular ways to generate complex business apps with a fraction of the code required of a 3GL, such as Java or C++. Pattern-based development methods, whereby developers use drag-and-drop GUIs to fine-tune pre-built components to create a customized business application, have been well-established in the IT industry for some time, although they have their share of detractors.
4. In AI, the tools and technologies for building machine learning models and deploying them into production are quite different than in general app dev, but the same low-code techniques still apply.
5. Low-code for AI allows citizen developers and data scientists to utilize AI building blocks to generate an AI engine that fits their needs. It puts ‘AI superpowers’ in the hands of users, eliminating the need to write complex code, compile, deploy, and scale it.
6. Organizations are realizing the value of AI in the post-pandemic world, and are also recognizing that they can’t afford to build AI solutions from the ground up due to the long deployment cycles.
7. Driven by a mission to change their operations faster than ever before, organizations are quickly adopting low code and AI to allow creative, mission-driven employees to innovate, regardless of their role and depth of technical expertise.
EXAMPLES
1. So, where can organizations get low-code tools for AI? Veritone develops an AI platform called aiWARE that’s designed to help companies use machine learning techniques to automate decision-making based on audio, video, and text input.
2. Another low-code platform gaining traction in the big data space is PyCaret, which is a Python library designed to enable users to perform complex machine learning tasks with just a few lines of code. The software, which debuted in April, incorporates other libraries and frameworks, including scikit-learn and XGBoost, and works in a notebook-style interface.
3. In October 2019, Qeexo released its AutoML offering as a stand-alone software offering. The product automates many steps in the ML process, from building models from collected data, comparing performance of those models, and then deploying the finished model to a microcontroller – all without requiring the user to write any code.
4. The offering has built-in support for the most popular ML algorithms, including random forest, gradient boosted machine, and linear regression, among others. Users can also select deep learning models, like convolutional neural networks, but many microcontrollers lack the memory to handle those libraries, Lee says.
5. Qeexo’s AutoML solution automatically handles many of the engineering tasks that would otherwise require the skills of a highly trained ML engineer, including feature selection and hyperparameter optimization. These feature are built into the Qeexo offering, which also sports a built-in C compiler and generates binary code that can be deployed to microcontrollers, such as those from Renasas Electronics.
AUTO- ML
1. You can also think of emerging AutoML tools as a form of low-code. These systems can automate many of the AI tasks that are traditionally handled by the data scientists, including model selection, parameter tuning, deployment into production, and post-deployment model management.
2. Reusability is a major benefit for today’s complex neural networks, such as ResNet, AlexNet, and GPT-3, the massive language generator from OpenAI that features a staggering 175 billion parameters. Data scientist can easily take these pre-built models, chop off the parts they don’t need, and retrain the remainder to do what they need. It may not be no-code, but the data scientists is certainly writing less code than if she were to start from scratch.
3. Similarly, many of the self-service mechanisms in today’s popular BI and visualization tools are helping analysts and others swing above their weight. As these BI tools are increasingly fused with ML capabilities, the impact of delivering advanced analytic capabilities in a self-service (i.e. low-code) manner can begin to have a real impact on data projects.
4. At the end of the day, low-code isn’t anything new, in AI or anywhere else. Not all AI projects will be amenable to low-code techniques. After all, the most compelling business transformations are almost always one-off projects that require many months of trial and error. What’s more, many real-world production AI systems will require the speed and performance that comes with optimized code, and that almost always has to be written by hand.
5. But for many other projects, there’s no contesting the real impact that organizations can find from allowing less experienced members of a team get more data work done in a quicker amount of time, particularly for less challenging AI projects and prototyping. Whatever you want to call it–citizen data science, AutoML, self-service BI, or low-code AI–that trend of doing more with less will continue.
https://www.datanami.com/2020/09/02/low-code-can-lower-the-barrier-to-entry-for-ai/
