The Future Stack AI Newsletter #5
Welcome to the fifth edition of the newsletter that brings you the latest developments in AI from all over the world.
⚡New this week
💗Mental Health Check-in
Take a moment to appreciate your loved ones, for they are the ones who make all the hard work, all the hustle, and basically everything worth it! 🤍
👩💻AI for practitioners
💸💸💸How to reduce the operational costs of your neural networks?
You might have been unlucky enough to encounter the dread of justifying the operational cost of an AI model. If you want the company to keep your model running, you need to make a business case for it, to them, and to yourself.
The deep learning community has also been exploring optimization techniques to help models run faster in the last few years. Why is optimization important? Well apart from reducing operational costs by reducing memory usage and the number of floating point operations, model optimization also allows for an increased model throughput. A combination of all these factors will result in highly efficient and cost-effective AI models that in turn make it easier for a higher RoI for your business problem. Imagine if you could cut your AI operational costs in half. That would look great to your boss come appraisal time 😉
The major optimization techniques used are network quantization, pruning, and, knowledge-distillation. But the problem with all these techniques is that the adoption of these techniques has been slow. All of these techniques have some trade-offs and have mostly been explored in research settings only.
NeuralMagic is a company in the US that is working on leveraging sparse AI models and the specific architecture of a CPU to run inference on CPUs faster than GPUs. They are basically democratizing sparse inferencing and making it much easier for anyone to leverage state-of-the-art optimization for their workflows. They apply a combination of pruning (to reduce the number of computations) and depth-first execution of a neural network to run inferencing efficiently on a CPU. They also have an open-source library you can try out. 🔗 Link here.
💡💡💡Dominate that conversation on Multimodal Fusion
Okay, first of all, what is multimodal data? This is data that comes in multiple modalities - text, images, videos, data from different sensors, data from different satellites having different temporal and spatial resolutions, etc.
Why is multimodal fusion valuable? AI tasks are becoming more and more complex and people want to do more and more with AI. But, an AI problem can only get as complex as the data it has. Enter multi-modal data fusion.
We cannot always get all the information we need about something from a single modality. The reason multimodal fusion is so powerful is that one modality can fill the gaps of the other and give us a much better picture than we would get with just one. Think about it like this - you are creating better representations of your data by combining different data sources and getting a feature vector that has all the information in it.
This finds application in agriculture, applications involving scene understanding and sentiment analysis, self-driving cars, healthcare including alzheimer’s, skin cancer, breast cancer prediction, and many other domains.
A few papers and links to get started:
🔗Multimodal deep learning GitHub
🔗Data Fusion in Agriculture
🔗Survey on multimodal sensor fusion for auto driving
🔗Multi-modal fusion transformer for E2E Autonomous Driving
🔗Fusing electronic health records and CT Scan for Pulmonary Embolism
🔥AI for business leaders
📈📈📈The inevitability of AI in an efficient economy
The idea of economic efficiency according to Investopedia is,
Economic efficiency is when all goods and factors of production in an economy are distributed or allocated to their most valuable uses and waste is eliminated or minimized.
The reason AI brings efficiency to industries is it helps makes decisions around efficient production, distribution, allocation, and all kinds of business processes.
As a business leader, where to improve efficiency should not come from assuming AI can help in all areas. It should come from your business’s data. Data always comes before AI, and data collection is the most important step.
Some common pitfalls even experienced business leaders fall into when applying AI in a new business context are:
Hiring a Machine Learning Engineer before even thinking about data governance
Not listening to your SME (Subject Matter Expert)
Applying AI just because your competitor is doing so or because everyone is talking about how great it is
Not listening to your data and trying to apply fancy techniques without looking at the data
Not defining metrics early on
Not knowing good data from bad data
Not caring about data governance
To make sure you don’t fall into this, make sure to talk to SMEs and data scientists and come up with a business goal that AI can help you target.
💯💯💯AI business idea of the week
A SaaS platform for web designers and developers that just lets you type in how you want your web page to look and makes it for you. The input can also be around style, not just color. Such plugins exist in Figma but they are not very easy to use, yet 😉
Hot tip: Use GPT-3 for this.
🕴️AI for investors
🧐🧐🧐What investors should ask before investing in AI companies?
Just because someone says they are using AI, does not mean that they are. When you are being pitched by a business saying that AI is at the core of their business, evaluating the claim thoroughly is critical. Here are a few questions that may help.
Does your product get better if more people use it?
This tells you whether the product gets better with more data. It also indirectly tells you two more things - have they implemented or thought about implementing a feedback loop and whether they have a way of distinguishing bad data from good data and training models on the good quality datasets?
This one question can tell you whether the business will get better with data learning effects which I have talked about previously as well.
🔗Click here to learn more about data learning effects.Who are the domain experts you hired?
This tells you if the company has a domain expert in the industry they serve. It says a lot about the depth in which the company has researched before building a product.What is the cost of a wrong prediction? Who bears this cost and how do you identify it?
When AI was introduced in the banking industry, one of the first processes to get automated or semi-automated was loan approval. It was all well and good till cracks started appearing in the predictions. The predictions put minority groups and low-income households at a disadvantage.
The problem is not just bias, there are algorithms that can reduce bias. The problem is that minorities have less data, and the predictions that any algorithm makes will always be less precise than the majority population.
🔗Read more here in MIT Technology Review
🔗Read the paper hereCompanies that have AI at the core of their business need to be aware of such issues, otherwise, the benefits might not be enough to justify using AI.
🦘AI for the Australian industry
🌲🌲🌲AI and climate change - what is going to be Australia’s role?
ACCESS-NRI in Australia is an earth-system simulator that is created to model past, present, and future climate and weather. AI can not only bolster simulations but also help create better what-if scenarios and show the impacts of extreme weather in a traceable and reproducible way.
ANU and InSpace-backed OzFuel can help find out the biomass content of forests and hence model the carbon sequestration of forests which can help quantify exactly how much carbon dioxide will need to be removed from the atmosphere.
Australian company MCi Carbon is speeding up the mineral carbonation process that occurs over millions of years to minutes and is “trapping” carbon dioxide in industrial building products.
In my opinion, Australia is going to play a major role in helping the world fight climate change, and this is a good time to bring together stakeholders including climate researchers and AI engineers to work on the most important problem that threatens humanity. AI is not a silver bullet, but definitely a technology that can help us speed up the mitigation process.
🔗Read how we can tackle climate change with machine learning
🔗Read what National Geographic commented on AI and climate change
🔗A BCG report on AI helping climate change
💰💰💰These Australian AI startups raised $142 million in Nov 2022
🔗Advanced Navigation: $108 million
Advanced Navigation specializes in the development of navigation technologies and robotics.
🔗Sapia: $19 million
Sapia is humanizing the recruiting process with AI smart interviewer, an automated talent solution for both candidates and employers.
🔗Abyss Solutions: $15 million
Autonomous inspection systems that leverage machine learning and robotics to deliver comprehensive inspections of critical infrastructure
🏭🏭🏭National Reconstruction Fund to drive sovereign capability
Australian Industry Minister Ed Husic announced the off-budget National Reconstruction Fund which will be able to pump $15 billion into the Australian manufacturing industry.
The government has identified seven priority areas: value-add in resources; agriculture, forestry and fisheries; transport; medical science; renewables and low-emission technologies; defence capability; and enabling capabilities such as fintech, artificial intelligence, robotics, and quantum physics.
This fund will make up for the post-pandemic VC dearth that Australia has been going through.
🔗Click here to know more about the NRF.
🔗Click here to know more about investment vehicles through NRF.
🔧AI for Talent
☁️☁️☁️As an ML Engineer, should you learn all 3 cloud platforms - Azure, GCP, and AWS?
I was of the strong opinion that you only need to invest time in one, but the more I work in the industry, the more I see that the best ML Engineers have deep experience in one, but know a lot about the rest of the cloud platforms too.
I specialize in AWS but I am currently doing a course in Google Cloud to learn more about Vertex AI. Will keep you updated!!
👩💻 About me
I am an ML engineer in Canberra and my most intense obsession is to help companies and individuals become competitive by investing in data and AI. Subscribe to my newsletter if you have similar ideas on making a dent in the universe. All the content that I create is free and I intend to keep it so!
Follow me on LinkedIn here!