Machine Learning with Images for ALL

Here is a write up about machine learning with images and a great Webinar run by TUANZ recently.

Unsupervised machine learning  has become highly influential in our lives and needs to be more widely understood.

Here are some descriptions of some of the  key concepts involved, discussion about the implications of its use, plus commentary about where the field could be going for more widespread deployment for community good.

I hope is useful and  it serves as a call to action  to encourage as wide a group of people as possible to imagine and create uses for unsupervised machine learning analysis, especially with images.

Unsupervised machine learning analysis has been useful for several decades, however, until recent times its use has been in the hands of big companies and organisations,  and it was only able to be configured by highly qualified experts.

Now, thanks to open source software, cheaper hardware and most important of all due to rapidly evolving  generative AI machine learning algorithms  its use is becoming accessible to a  much wider range of people who are prepared to take the time to learn the concepts and set up systems to achieve all manner of useful things.

“Unsupervised machine learning uses machine learning algorithms to analyze and cluster unlabelled datasets (subsets called clusters). These algorithms discover hidden patterns or data groupings without the need for human intervention. This method’s ability to discover similarities and differences in information make it ideal for exploratory data analysis, cross-selling strategies, customer segmentation, and image and pattern recognition.” From: What is machine learning (ML)?,and%20image%20and%20pattern%20recognition.

Now our computers and laptops are increasingly being built with neural processor units (NPUs) and relatively cheap server computers are available with NPUs to enable machine learning to be done by ‘everyone’.

“Neural networks (also known as artificial neural networks or neural nets, abbreviated ANN or NN) are a branch of machine learning models inspired by the neuronal organisation found in the biological neural networks in animal brains.” From:  Neural network (machine learning)

One of the significant challenges in deploying machine learning solutions is the substantial concern surrounding data privacy. As ML models require vast amounts of data to learn and make predictions, the collection and utilisation of such data raise critical privacy issues, especially when dealing with sensitive information. The need to ensure that data is collected, stored, and processed in a manner that complies with stringent data protection regulations, like the General Data Protection Regulation (GDPR) in Europe, can be a complex and resource-intensive task. Furthermore, there’s the risk of unintended data exposure through ML models themselves, known as data leakage, where sensitive information could potentially be inferred from a model’s output.


Another core limitation is the necessity for large, high-quality datasets to train models effectively. The adage “garbage in, garbage out” holds particularly true in the context of machine learning, where the quality of the input data directly impacts the performance and reliability of the resulting model. This requirement for substantial amounts of good quality data poses a barrier, especially for organisations lacking access to such resources. Additionally, there’s the issue of bias in AI, which stems from biased data sets or flawed assumptions made during the model development process. Bias in machine learning can lead to skewed outcomes that might unfairly disadvantage certain groups, perpetuating or even exacerbating existing societal inequalities. Addressing these biases requires diligent efforts in dataset curation, model training, and ongoing monitoring to ensure equitable outcomes.


These challenges underscore the importance of a meticulous approach in the development and deployment of machine learning models, emphasizing the need for robust data governance, ethical AI practices, and continuous efforts to mitigate bias, ensuring the responsible use of technology.

From Concept to Capture: Navigating Real-World Machine Learning for Image Analysis

I attended a great TUANZ Webinar last week about Machine Learning  with Images. How machine learning analysing images can be relatively easily done using an apparently easy to obtain server with a neural processor unit (NPU),  running on open source Software from the users premises. Evidently this costs under $2,000.  They work on exponentially less power than a PC too, a NPC server uses 9 watts versus 600-700 watts for a PC.

Craig Young, TUANZ CEO interviewed Auckland-based entrepreneur Mike Milne, Kauricone  about his amazing work using one of these servers to generate systems  to help address local issues and problems.

Mike used three examples of his work:

  1. Recognising and measuring salvageable frame timber ‘rubbish’ amongst other demolition materials in skips on building sites and calculating the salvageable value.

  2. Detecting theft in a car park building being committed by thieves on scooters able to move with agility and speed.  Mike’s AI agent was trained to recognise them, via the camera’s installed around the car park building, as soon as the AI system recognised a  theft occurring, or a suspicious activity it activated an alert to the security people for the building.

  3. Recognising rockfalls on the road.  Solar-powered cameras are placed along sections of roads  known for dangerous rockfalls.  Images of the roadway sections are taken frequently and fed into Mike’s AI system which has been trained to recognise any rockfalls and can alert road engineers to them almost immediately.

Here is the YouTube video Craig has posted of the Webinar:

Call To Action

This is the sort of thing that could drive re-establishing or strengthening community creativity hubs generating ideas for local uses.. I bet places like the Blueprint in Palmerston North will do this, maybe they already are.

Great to see University of Auckland has established a Centre of Machine Learning for Social Good

“The mission of the Centre of Machine Learning for Social Good is to advance fundamental knowledge in machine learning and data analytics while addressing the most challenging and pressing health, environmental, and societal problems of our time. This is the first centre in Aotearoa focussing on social good by using machine learning in collaboration with domain experts as a catalyst to solve high-impact societal issues.”

I am sure centres like this will be ever increasingly springing up to assist us – you can be sure I’ll be encouraging getting a centre is established in my home district.

For those interested in diving into machine learning, especially in the context of image analysis, there are several accessible resources to kickstart your learning journey. Whether you’re a complete beginner or have some knowledge and want to deepen your understanding, these platforms offer structured pathways and comprehensive insights into machine learning principles, applications, and hands-on practices.

  1. TensorFlow – A robust resource for beginners and advanced learners alike, offering everything from introductions to machine learning basics to in-depth tutorials on TensorFlow, a leading machine learning library. Notably, it provides a quickstart guide for beginners focused on using Keras for image classification, and various courses and books for deepening your understanding of machine learning, including “Hands-on Machine Learning with Scikit-Learn, Keras, and TensorFlow” by Aurélien Géron and “Deep Learning” by Ian Goodfellow, Yoshua Bengio, and Aaron Courville. TensorFlow also lists online courses and specializations such as the TensorFlow Developer Specialization on Coursera, designed to teach best practices for using TensorFlow​​.
  2. Springboard – Offers a comprehensive list of 40 free resources to help you learn machine learning on your own. This includes a well-structured path for learning Python, vital for machine learning, alongside introductions to crucial mathematical concepts through courses like “Data Science Math Skills” by Coursera. It covers machine learning basics like probability theory and statistical error, and guides on data pipelines, probabilistic programming, and understanding machine learning algorithms. It also encourages applying what you’ve learned through practice with various case studies and data sets from platforms like Kaggle and GitHub​​.
  3. DataCamp – Presents an approachable guide on image analysis within machine learning. It offers a beginner-friendly explanation of morphological operations, including dilation and erosion, which are key techniques in image preprocessing. Moreover, DataCamp delves into supervised machine learning for image analysis, covering topics from training models to practical applications like image classification and object detection. It emphasizes the use of convolutional neural networks (CNNs) for feature extraction and provides insights into the architecture and training process of these models​​.

These resources not only offer theoretical knowledge but also hands-on projects and examples that allow you to apply what you’ve learned in practical scenarios. By leveraging these platforms, you can build a solid foundation in machine learning and image analysis, paving the way for advanced studies or careers in this dynamic field.

In another blog soon I intend to look at using machine learning using intelligent agents.  It  looks like these will be the way we can get more personally involved in designing uses for them.

Intelligent Agent :


This blog post is a collaborative creation by Alistair Fraser, with the innovative assistance of OpenAI’s ChatGPT-4 and DALL-E 3, showcasing the synergy of human creativity and advanced AI technology.

Leave a Comment

Your email address will not be published. Required fields are marked *

Discover more from Helping to Achieve Our Dreams

Subscribe now to keep reading and get access to the full archive.

Continue reading

Scroll to Top