วีดีโอ

It is possible to get the python scripts

Thanks for sharing. Nice testimonial.

Thanks for sharing, Nice testimonial.

Thanks for sharing. Nice testimonial.

Thanks for sharing

Great session and practical insight into this world by Dr Satnam. Can we get those references and specially the mind map of learning path( if it posted in Sabudh repo somewhere?)

Good sir

A very informative session!!

A very informative session!!

Thank you Sabudh for quality content.

Excellent talk. Thank you.

The QnA session was wonderful.

Status paga

Could you please tell how to grab an internship in data science in sabudh foundation

What is the distribution of random vector d? Is that always fixed? Or time-varying?

Wonderful Talk! Thanks! I have a question from slide 6, do you have any mathematical explanation/framework of the terminologies ''prediction'' and ''trust'' in this context?

Now i see why they always recommending #haxtechie01 on IG you are the main boss thanks for the successful Unlock

Thank you for doing such a great seminar!

Thanks for sharing the recording here.

Big thank you Sarab sir for conducting these webinars. These are eye opening use cases and make us aware about the depth of great work going around on this subject. Would love to listen to your journey as well as you have experienced alot of these advances in this technological domain. Many thanks!

Challenges for ML at scale: 1. Fairness Survey hindi -> data biased -> algo inherent bias -> discriminated outputs May'19 GDPR by EU. We don't know if those bias variables exists in dataset. California vs NY gender vs car color - correlated. Race-income Race-education How algo can be less discriminatory. Ethical data modelling. Use less demographic variables Ai fairness 360 degrees toolkit detects bias variables 2. Reliability How robust are the results. DL ability to predict with higher accuracy has risen 2012 onwards Model shd be generalized enough Training algo robustness Adversarial impact /attacks Panda image + noise=junk Have enough training representation, so as to easily side line adversarial data. More unstructured data today. Comf matrix Softmax func 3. Explainability Explainable AI ML is not a blackbox, we shd be able to explain the reason for outcome. Answerable to auditor. FAT US millitary research wing XAI explainable AI Writting letters -> automatic classification based on pincodes 85-90% accurate Friend past 5 yrs pics, can we predict what next? Hill station predicted Sea beach actual How? Which variable is impacting model predictions? LIME LOCAL INTERPRETABILITY LOCO leave one covarient out ELI5 Explain like I am 5 follows LIME SHAP sharply Debugging model predictions becomes easy. Helps in accountability of our model. Investment -> take decisions -> a risk Can you pls share email resources?? Accuracy Fairness training set Reliability attacks Explainability key decision makers Textual data Is GDPR truly achievable? Not fully. What is sensitive info and what is not? Behavioral data? Customer centric data? Personally identifiable data? Information asymmetry Drivers info Mostly male drivers Gender for female can be biased Use less survey driven data Survey vs actual huge gap