What is Kurtosis? Can we teach common sense to AI?

Hi, this is Data Pragmatist with another free issue of the Newsletter tailored specifically for you. We are on a mission to make staying up-to-date with the world of data and AI easier. Share with your friends and Colleagues.

A new day calls for a new edition, and put on your glasses as today you will learn some new things in the next 5 minutes. I am excited to share a lesser-known statistical concept and the trick needed to teach AI some sense.

Kurtosis: Understanding the Shape of Data Peaks

The name sounds like a Russian term, but the concept isn’t as tough as it sounds. It’s a simple concept that will help you understand the shape of data peaks in a data set. Kurtosis is a measure of the tailedness of a distribution. Tailedness is how often outliers occur.

Now, imagine a manufacturing unit producing bottles or spare parts, the team has to monitor the consistency of the product attributes, like the weight, dimensions and quality of the product. It is difficult to identify anomalies manually, without making an error. In this case, the Kurtosis of relevant measurements, such as weights or dimensions can help you identify anomalies and manufacturing defaults. How swift is that?

Kurtosis helps us identify consistency in the results, like consistency of weights or returns etc. In data science, kurtosis helps us decide which statistical methods to use. If data has high kurtosis, it might be better to use methods suited for data with a strong central peak. If data has low kurtosis, methods used for more spread-out data could be more appropriate.

If that makes sense, let’s learn more about how to make AI get some Common Sense.

Make AI make sense: Enhancing Common Sense in AI

Mankind has been obsessed with replicating thinking and the process of the human brain. Researchers have been working on deep neural networks to make natural language processing and autonomous image recognition possible. Now that it has been attained, it is time to enhance its accuracy. Neuro-symbolic models help you achieve it with high accuracy compared to other models, with less training data.

What is Neuro Symbolic AI?

Neuro Symbolic AI is an interdisciplinary field that combines neural networks, which are part of deep learning, with symbolic reasoning techniques. It aims to bridge the gap between symbolic reasoning and statistical learning by integrating the strengths of both approaches. This hybrid approach enables machines to reason symbolically, while also leveraging the powerful pattern-recognition capabilities of neural networks.

For example, currently, if an AI can identify the shape and colour of an object, symbolic reasoning can enable it to reveal its area, volume etc.,

Need for Neuro-Symbolic AI

Imagine trying to make robots think, and reason like humans do. Turns out symbols play a big role in how we communicate and think, making them super important for our smarts. People tried giving robots rules to follow, like a game plan, which worked but had some issues. It made the robot programs expensive and less accurate as they piled on more and more rules.

So, researchers came up with a new idea. They mixed two things: symbol stuff and brainy networks called neural networks. These networks can learn by themselves if they have lots of examples. But, hold up, there's a catch. These networks need tons of data and can be hard to understand.

Here's the cool part: they combined these two things to make something even better called neuro-symbolic AI. It's like having a super team-up of brainy network skills and clever rule-following. It's like the robots are getting the best of both worlds!

The plan is to make the robots even smarter by putting logic and learning together. See, the brainy networks are awesome at spotting patterns in a huge pile of stuff, while the rule-following side is like a super brain for understanding tricky things.

Imagine a robot using both its brainy network skills and clever rule-following to answer questions, just like a human. But guess what? It needs way less training than the brainy networks alone. So, with neuro-symbolic AI, robots can be super brainy and super smart at the same time!

Advantages of Neuro-Symbolic AI

• Higher Accuracy - Normal Neural Networks have already achieved 80% accuracy, then why bother about the 20%? Well, this is the part that makes us human. This part acts when no rules apply, like when a traffic signal is burning, should a car stop or move? The AI can’t decide anything in this scenario.

• Data Efficiency - Feeding Huge amounts of data into The AI system to train is time and cost-consuming. A Neuro-symbolic AI system learns from a few examples like the human brain, and hence the data fed is just 1% of what was required earlier.

• Transparency and Interpretability - AI can make decisions, and certain parts of its decision as to why it was made, remain unexplainable, even by the engineers. Neuro-Symbolic AI makes the decision-making process transparent and interpretable by artificial intelligence engineers.

In conclusion, neuro-symbolic AI is a promising field that aims to integrate the strengths of both neural networks and symbolic reasoning to form a hybrid architecture capable of performing a wider range of tasks than either component alone. With its combination of deep learning and logical inference, neuro-symbolic AI has the potential to revolutionize the way we interact with and understand AI systems.

If AI is getting smarter, so should you. Here is a pro tip to make your Data Visualisation Stand-Out.

📊 Choosing the perfect data visualization model for your datasets is like picking the right tool for the job! 🛠️ Here are some quick tips to help you nail it:

1. Know Your Data: Is it numbers or categories? Get this right from the start!

2. Show Patterns: Want to highlight trends? Line charts and scatter plots got your back!

3. Size Matters: Big data? Summarize with histograms or heat maps. Small data? Go detailed with scatter plots!

4. Group Comparisons: Group showdown? Stacked bar charts or side-by-side box plots are your pals.

5. Location Matters: Map it out with maps or choropleth maps for that geographic touch.

6. Time Travel: For time trends, line charts or area charts are your go-to pals.

7. Complex Stuff: Multivariate? Parallel coordinate plots or bubble charts can do the magic.

8. Spotting Outliers: Unusual data? Box plots and scatter plots are your detective tools.

9. Story for Everyone: Think about your audience and make sure your choice tells the story right.

10. Interaction Fun: Want users to explore? Go for interactive tools that zoom and filter.

11. Looks Matter: Keep it clean and classy. Choose colours that pop but don't hurt the eyes.

12. Make It Accessible: Everyone in? Use labels and alt text for all to enjoy.

13. Tools of Choice: Excel, Tableau, D3.js - pick what you're comfy with!

14. Experiment & Evolve: Don't be shy to try different visuals. It's a journey!

Get ready to rock your data and share insights like a pro! Let’s meet again soon with more interesting content.

Follow us on social media for more up-to-date content about data trends, ML and AI.

🐦 Twitter: @DataPragmatist

💼 LinkedIn DataPragmatist

This post is public, so feel free to share and forward it.

Read