Best Linkedin posts about Machine Learning

Don't use machine learning when • customer uses fax 📠 - they are decades behind in tech and mindset. • data origin is mysterious 🤔 - unclear what the model learns . • task has an unsolved social component 🤝 - ML won't magically fix it.

My big takeaway from Google's machine learning practice... Don't let perfection get in the way of good enough. Unless human lives depend on it, the sooner you put your machine learning model into production, the better. The longer you wait, the longer it will take to tackle issues in the wild. The longer you wait, the more the data potentially drifts. The longer you wait, the more leadership grows impatient. It's not going to be perfect regardless of when you launch, so the sooner the better. Find out what's "good enough" and just go for it. Love to hear from you if you disagree and why. See you in the comments ✌🏼. #machinelearning #datascience

Machine Learning is not about the fitting best model on data. The real-life problems aren't like the college projects where all the focus is on algorithms. In real life, you deal with many things related to: --> How will the data be collected? --> How will you handle such large data? --> How will you efficiently train the model? [Keeping time in consideration] --> How will you use the prediction result to solve the problem? --> How will we optimize the cost for all these? And most importantly, how will you explain to the non-technical person that ML models don't have 100% accuracy? Yes, the algorithm is part to consider, but the technical person like ML engineer/Data Scientist bothers more and not everyone. Solve problems, and be happy. Don't bother to make everyone understand your solution. #datascientist #machinelearning #ai #datascience

Please STOP making this critical mistake when working on actual industry projects in data science. I have made this mistake too. Do NOT pay excessive importance to the test data performance for a machine learning model. Why? - Test data is just static unseen data which tries to be a simulation of real-world data - Often real-world data may have changed vs. the initial data you use for model training and evaluation Solution: - Deploy a minimal model training and inference pipeline as soon as you have an initial working model and start evaluating and testing on actual data - Build feedback loops and then keep improving and switching out models as needed Books and courses are often too academic and place undue importance on test data but please don't make it the only way to evaluate your ML models. Hope this helps! #machinelearning #data #datascience #mlops #analytics #statistics #artificialintelligence #ai

🟢 Here's exactly what you should know about machine learning! #machinelearning #datascience

How to build a bullet-proof machine learning system in 2022: 1. Have a "plan B". It doesn't matter how much you tested your model, things can break in production. Built experimentation infrastructure and use A/B test to roll out the new model slowly, and if it triggers bugs, roll it back to the 'control'. Putting models in production should never be a 'one-way' door. 2. Plan your infrastructure for 10X growth. If you worry about the capacity, plan the overhead. It might not be 10X in your case, but you need to really understand the logs and metrics you collected from experimentations to estimate the patterns in traffic. Always leave extra room in case of black swan events. 3. Don't 'overfit' your system. If your machine learning pipeline becomes very specific to one type of model, it's hard for data scientists to innovate. If you can build a foundational system that data scientists from different teams can easily test and launch models, it'll save more engineering hours compared to perfecting the system just for one use case that made you start this pipeline. Machine learning models constantly evolve, and you should leave room for changes. A lot of teams are still early in the ML journey, and instead of avoiding failure, designing the system to handle risks so we can safely innovate. This is inspired by my chat with 👩🏻‍💻 Mikiko B., learn more from her on "the data scientist show". Links in comments. What principles do you follow when it comes to building ML systems? #machinelearning #datascience

When people say "They're a good data scientist" What they mean: - They collaborate well with other data professionals - They know when to reach for machine learning - They build solutions that can be readily tested - They seek to deeply understand the user case - They link their work to the value delivered - They deeply understand the technology - They get regular feedback from SMEs - They aim for production at the outset - They prioritise simple over complex - They seek small wins - They iterate quickly Anything missed? #DataScience #Technology #MachineLearning P.S. I wrote about this in my newsletter today link below 👇

Alert: Harsh Sarcasm ahead! Read at your own risk. Dear friends, Please don't write to me and ask me to be your mentor 🙏🏻. I'm yet to do many things in Machine Learning and AI. At most I can 'guide' you on what to do and more importantly what not to do. The day I think I'm useless and have nothing to do I will update my LinkedIn profile and add 'Mentor' to it 😜 . Until then I would like to solve real-life problems! PS: If this offends you, then I'm sorry, but then it also means that it was written for you! #sarcasm #python #machinelearning #statistics

My phases of working with machine learning as a data scientist. Phase 1... This ML stuff is so cool! But how does it even work? Phase 2... I want to learn everything possible about machine learning! Phase 3... I only want to focus on deep learning! Let's try recreating everything from scratch. Phase 4... This random forest stuff is actually really good. Why don't more people talk about this? Now... Let AutoML figure out which algorithm to use. Let's focus on the data and the business. Have you gone through something like this? Which phase are you in? See you in the comments ✌🏼. #machinelearning #datascience

Is Google helping spammers become better? Looking through my Gmail this morning and realized that it gives you labelled spam and not spam emails. (Yes, I understand they show the spam emails in case something's been misclassified.) But couldn't you just train a machine learning model using the labelled data from Gmail to identify spam vs. not? And then use that to fool Gmail's spam filter? Am I crazy to think that machine learning model could beat Gmail's spam filter? Thoughts? #machinelearning #datascience

AI for Engineers 👇 Beginner: I just want to use Machine Learning💡 Intermediate: I have data and want to use Machine Learning 📊 Advanced: I have high quality data and know the problems I’m trying to solve 📈 Master: What’s the business value? 🧠 #machinelearning #engineers #business #data #data #ai #quality

Every machine learning course talks about splitting your data. Surprisingly, many people don't understand how to use each set properly. Let's talk about some of the things you should know about splitting your data. We usually split the data into three different sets: 1. Train set 2. Validation set 3. Test set The first thing you should do at this point: Forget that your test set exists. Train set: The data you'll use to train your model. This set is your entire world. No other data exists outside of this train set. Going forward, you'll use this data for every analysis, transformation, and decision. Validation set: As you experiment, you'll use this data to compute your model's performance and decide how to improve it. The validation set gives you feedback. You can use this feedback to improve your model. Here is the iterative process we follow: • Train a model • Evaluate it with your validation set • Improve the model • Evaluate it with your validation set • Improve the model Because of this process, your model will start overfitting to the validation set after some time. It will become good at predicting the validation data, which won't be helpful anymore. That sucks, but here is how you fix it: After several iterations, throw your validation set into your train set and get new data. If you don't have more data, you must rely heavily on k-fold cross-validation. Test set: Until the very end, you never look at your test data. You never use it to do any analysis or transformations. Never make decisions that affect your model using the test data. You treat your test data as if it doesn't exist. The goal of your test set: To provide a final, unbiased estimation of your model's performance. A good test set will give you a performance similar to what you expect when processing production data. Many people run their model on their test set and discover that their model is not good. They go back and make changes to the model until the performance improves. Nothing wrong, except when they use the same test set again! Use your test data once. After that, merge it into your train set and find new test data. The effectiveness of your test set decreases proportionally with the number of times you use it. Soon the test set won't longer be an accurate measure of how good your model is. Let me TL;DR this quickly: 1. After a few iterations, rotate your validation data. 2. Don't use your test set more than once. #machinelearning #datascience

What is Machine Learning? In the simplest definition, #machinelearning is fitting a function to examples and then using that function to generalize and make predictions about new examples. It is based on the data that you feed it. It learns from examples. And the entire goal is to use that learned model to make predictions about new examples.  In other words, machine learning models learn from the trends in past data and then try to find those trends in future data to make predictions. A simple example is a fraud detection model for a credit card company. The model learns what conditions typically indicate a fraudulent charge based on past data, and then when it's presented with a new charge that fits those conditions, it will predict that the charge is fraudulent. Agree? Do you want to add any more definitions and examples? Please let's talk in the comments 👇 #datascientist #analytics #data #unitedarabemirates  #artificialintelligence -------- 🔔 Want more content like this in your LinkedIn feed? Then don't forget to Follow Mohammad Arshad 👉🏻My mentoring and newsletter link is in the comment below.