Yashika Sharma

into AI-ML & Quants

[Yashika Sharma @Yaashix97]

Funny how one small change can completely change the outcome.
In deep learning, a tiny tweak to the learning rate can change everything.
Maybe life works the same way.

[Yashika Sharma @Yaashix97]

Funny how one small change can completely change the outcome.
In deep learning, a tiny tweak to the learning rate can change everything.
Maybe life works the same way.

[Yashika Sharma @Yaashix97]

Day 08/30 -> Deep Learning

Today's focus was on making neural networks train more efficiently rather than just making them bigger.

I learned how Batch, Stochastic, and Mini-batch Gradient Descent differ
why vectorization speeds up training
what happens when the batch size doesn't evenly divide the dataset and
how hyperparameter tuning can make a surprising difference.

Every day I realize that building a model is only half the job.

Training it well is where the real learning begins.

On to Day 09 ;)

[Yashika Sharma @Yaashix97]

Day 08/30 -> Deep Learning

Today's focus was on making neural networks train more efficiently rather than just making them bigger.

I learned how Batch, Stochastic, and Mini-batch Gradient Descent differ
why vectorization speeds up training
what happens when the batch size doesn't evenly divide the dataset and
how hyperparameter tuning can make a surprising difference.

Every day I realize that building a model is only half the job.

Training it well is where the real learning begins.

On to Day 09 ;)

[Yashika Sharma @Yaashix97]

@import_alis Thanks dude :)

[Yashika Sharma @Yaashix97]

Day 07/30 -> Deep Learning

Learned about MLP memorization today.

It was interesting to see that neural networks can sometimes memorize examples instead of learning actual patterns.

Good reminder that great training accuracy doesn't always mean a great model.
One more concept down, many more to go.

[Yashika Sharma @Yaashix97]

Day 07/30 -> Deep Learning

Learned about MLP memorization today.

It was interesting to see that neural networks can sometimes memorize examples instead of learning actual patterns.

Good reminder that great training accuracy doesn't always mean a great model.
One more concept down, many more to go.

[Yashika Sharma @Yaashix97]

Day 06/30 -Deep Learning
Today I learned Backpropagation->the algorithm that enables neural networks to learn from their mistakes.
Covered: • Why backpropagation is needed • How errors flow backward • Chain rule for gradient computation • Gradients of weights & biases • Gradient descent updates • Learning rate • The complete training loop

It's fascinating how a model improves simply by reducing its errors one step at a time.
"ll be back tomorrow:)

[Yashika Sharma @Yaashix97]

Day 06/30 -Deep Learning
Today I learned Backpropagation->the algorithm that enables neural networks to learn from their mistakes.
Covered: • Why backpropagation is needed • How errors flow backward • Chain rule for gradient computation • Gradients of weights & biases • Gradient descent updates • Learning rate • The complete training loop

It's fascinating how a model improves simply by reducing its errors one step at a time.
"ll be back tomorrow:)

[Yashika Sharma @Yaashix97]

Day 05/30 - Deep Learning.

Today's topic was all about understanding how a neural network actually works.

I learned how forward propagation helps the network make predictions, why we need loss functions, the difference between loss and cost functions, and the pros and cons of different loss functions.

The biggest takeaway for me was that a model doesn't improve on its own-it learns by comparing its predictions with the actual answer and gradually reducing its mistakes.

Still a long way to go, but enjoying the process. See you on Day 06.

[Yashika Sharma @Yaashix97]

Day 05/30 - Deep Learning.

Today's topic was all about understanding how a neural network actually works.

I learned how forward propagation helps the network make predictions, why we need loss functions, the difference between loss and cost functions, and the pros and cons of different loss functions.

The biggest takeaway for me was that a model doesn't improve on its own-it learns by comparing its predictions with the actual answer and gradually reducing its mistakes.

Still a long way to go, but enjoying the process. See you on Day 06.

[Yashika Sharma @Yaashix97]

Day 04/30 - Deep Learning

Spent today's study session understanding why a single perceptron has its limits.
Learned about MLP notation, why hidden layers are needed, what happens when we add more neurons to different layers, and how Deep Neural Networks are built.

It's interesting how just adding layers completely changes what a model can learn.

Back to studying. Day 05 next🙌

[Yashika Sharma @Yaashix97]

Day 04/30 - Deep Learning

Spent today's study session understanding why a single perceptron has its limits.
Learned about MLP notation, why hidden layers are needed, what happens when we add more neurons to different layers, and how Deep Neural Networks are built.

It's interesting how just adding layers completely changes what a model can learn.

Back to studying. Day 05 next🙌

[Yashika Sharma @Yaashix97]

The topics covered here are :

•How to take input and output in C++
•Datatypes
•Diff b/w int and long long int
•Constants
•Variables
•Operators( unary,arithmetic, conditional, relational, logical)
•Conditional statements(if else)

[Yashika Sharma @Yaashix97]

Day 01 notes of C++ which I didn't shared earlier.😊

[Yashika Sharma @Yaashix97]

Day 01 notes of C++ which I didn't shared earlier.😊

[Yashika Sharma @Yaashix97]

Day 03/30 - Deep Learning.
Today's session answered a question I didn't even know I had: How does a model know it's wrong?

Yesterday I learnt how a perceptron draws a decision boundary. Today I learnt what happens when that boundary isn't good enough.

Spent time understanding: • Why the perceptron has limitations • What a loss function actually tells us • Perceptron loss function • Gradient descent (this finally started making sense 😅) • Training a perceptron using sklearn • Different loss functions and where they're used
The biggest takeaway for me was that learning isn't about getting the right answer immediately. It's about making a mistake, measuring it, adjusting a little, and trying again. That's exactly what gradient descent does.

Slowly connecting the dots.
Will be back tomorrow:)

[Yashika Sharma @Yaashix97]

Day 03/30 - Deep Learning.
Today's session answered a question I didn't even know I had: How does a model know it's wrong?

Yesterday I learnt how a perceptron draws a decision boundary. Today I learnt what happens when that boundary isn't good enough.

Spent time understanding: • Why the perceptron has limitations • What a loss function actually tells us • Perceptron loss function • Gradient descent (this finally started making sense 😅) • Training a perceptron using sklearn • Different loss functions and where they're used
The biggest takeaway for me was that learning isn't about getting the right answer immediately. It's about making a mistake, measuring it, adjusting a little, and trying again. That's exactly what gradient descent does.

Slowly connecting the dots.
Will be back tomorrow:)

[Yashika Sharma @Yaashix97]

Day 02/30 – Deep Learning

Today's lecture was all about understanding the perceptron a little better.
• Learnt the perceptron trick
• Understood how decision boundaries split different regions
• Finally got why changing the weights rotates the decision boundary while the bias shifts it
• Studied feature transformations
• Implemented the perceptron learning algorithm
It feels good when concepts that looked confusing yesterday start making sense today.
Back again tomorrow :)

[Yashika Sharma @Yaashix97]

Day 02/30 – Deep Learning

Today's lecture was all about understanding the perceptron a little better.
• Learnt the perceptron trick
• Understood how decision boundaries split different regions
• Finally got why changing the weights rotates the decision boundary while the bias shifts it
• Studied feature transformations
• Implemented the perceptron learning algorithm
It feels good when concepts that looked confusing yesterday start making sense today.
Back again tomorrow :)