Sijan Bhandari

Learning

[WFP Nepal @WFP_Nepal]

📍Gorkha🇳🇵

In the high mountains of #Chumling, #Ranagaun & #Dorjung, women masons walk two hours everyday to work.👩‍🏭🚧

Here, @WFP reconstructed 8.5 km of trails, benefiting 1,844 ppl w/improved livelihoods & easy access to markets, schools & health posts.

Thank you, Germany🇩🇪.

[Sijan Bhandari @sijanonly]

@BHampson @DjangoConEurope Hi Ben, Happy to see you again. Thanks for connecting here. I am loving virtual conferences :)

[Sijan Bhandari @sijanonly]

Starting my Django conference with the first talk "Programming for pleasure" :)
@DjangoConEurope

[Prashant @capeandcode]

Precision and Recall are confusing initially🤔

The TP, TF notations don't do a favor while trying to understand them intuitively. I never could remember them until I understood the concept.

So let's talk about them in terms of picking apples and oranges🍎🍊, because why not! 👇

[DjangoCon Europe @DjangoConEurope]

🚨 TICKET GIVEAWAY 🚨
Retweet this tweet for a chance to win a #djangocon ticket!

ℹ️ DjangoCon Europe 2021
🌍 Online
🗓 2-6 June 2021

https://t.co/1EDgOz5GkX

#ticketgiveaway #django #python #conference

[Quiz on Machine Learning @ml_quiz]

Bayesian Inference can be summarized into 4 major steps:

a. Calculate Likelihood
b. Calculate/Collect prior
c. Calculate posterior
d. Inference.

Which one represents the correct order ?

#MachineLearning #DeepLearning #NeuralNetworks #ArtificialIntelligence #DataScience

[Quiz on Machine Learning @ml_quiz]

Given the data 'D' and the model (parameter) 'θ', which of the following is called likelihood ?

#MachineLearning #DeepLearning #NeuralNetworks #ArtificialIntelligence #DataScience #bayesianlearning #quiz

[Alexey Grigorev @Al_Grigor]

I have 5 copies of Machine Learning Bookcamp

To win one:

🔸 Follow me
🔸 Retweet this tweet

Winners selected randomly. Results announced on Wednesday

ML Bookcamp - Learn machine learning by doing projects

🔗 https://t.co/m4IrQc7hle

@ManningBooks

[Sijan Bhandari @sijanonly]

@haltakov Formulation :

b= the distance between two cameras
f= focal length of camera,
d= disparity:
D = Distance of point in real world,
->D = b*f/d

4/4

[Sijan Bhandari @sijanonly]

@haltakov 1. Both capture the images.
2. we will find the location of the front-car in both of the images.
-> we need object detection here.
3. We match the object in both the images (stereo matching)
4. compute the disparity measure
5. use disparity to estimate the distance.

3/4

[Vlad Haltakov @haltakov]

Machine Learning Interview Question #13 🤖🧠🧐

We talked about using CNNs to detect objects. Now imagine we have a self-driving car driving around.

❓ How can we estimate the distance to a detected object (for example a car)❓

Answer in the replies. Read the rules 👇

[Will Koehrsen @koehrsen_will]

Python default arguments are evaluated ONCE when the function is defined. This can have unexpected side effects:

[Sijan Bhandari @sijanonly]

@haltakov While formulating this solution, I just figured out the flaw :).

How to decide the patch size because objects may have different aspect ratios.

3/3

[Sijan Bhandari @sijanonly]

@haltakov So, in general, we can apply CNN with many patches of the given images and CNN should classify each patch as :
either object or maybe background and each patch itself gives the bounding box of the object.

2/3

[Vlad Haltakov @haltakov]

Machine Learning Interview Question #12 🤖🧠🧐

We talked about CNNs so let's use one of them now...

❓ How can you use a CNN to detect the presence and the location (bounding box) of a specific object in an image (for example car)?❓

Answer in the replies. Read the rules 👇

[Sijan Bhandari @sijanonly]

@chuston1776 @svpino At the same time, the loss function with smaller batch sizes will give unstable landscapes in each iteration. In this way, it is easier to escape those sharp minima.

4/4

[Sijan Bhandari @sijanonly]

@chuston1776 @svpino One intuition I have is - When it comes to using Large batch size, the shape of the loss function will remain consistent over the batches. This consistency is not good. Because this won't allow us to escape or so to say use other possible landscapes in the error surface.

3/4