๐ง Building cool things in public
๐ค Sharing AI, DSA & coding insights
๐ Making tech fun with humor + memes
๐ Follow for daily dev motivation & learning
Hey, Iโm Jupiter ๐
I tweet about:
๐ค AI & coding explained simply
๐ป My dev journey & DSA learnings
๐ Funny tech takes (because debugging is pain)
Follow if you like tech + humor ๐
I am on a Journey let's see where I go?
Insight's u might like:
Why does sunlight get 4x weaker when you double your distance from it?
Not because the Sun loses energy.
Because the same energy spreads over 4x the area.
That's the inverse square law
Physics explains it
Life reflects it
Data Science Reality Check
Train: 70% Test: 69%
At least itโs not overfitting!
Actually, itโs Underfitting
Model is stable, but consistently wrong
Itโs like a GPS that tells everyone "turn left" regardless of destination
Stable failure is still failure add some complexity
Stop guessing and start diagnosing:
Underfitting: Too simple. Fails everywhere
Overfitting: Too complex. Memorizes the past, fails the future.
Good Model: Generalizes well the Sweet Spot
If your Train and Test scores are both 70% don't celebrate model is just consistently wrong
Machine Learning just like dating
High Bias/Underfitting: You have one rigid "type" and ignore everyone else
Result: You're lonely
High Variance/Overfitting: You over-analyze everything from "Hey" vs "Hi"
Result: Total paralysis
Finding that sweet spot is basically a miracle
Underfitting ๐
Underfitting is when a model is too "lazy" to learn the actual pattern in your data
Both training and testing accuracy stay stubbornly low
It happens because the model is too simple to grasp the underlying complexity
#MachineLearning#AI
Fixing a Clingy Model ๐ค
If your model is overfitting, itโs just memorizing the answer key
Diet: Use a simpler model
The "Tax": L1/L2 Regularization
The Chaos: Dropout to keep neurons on their toes
The Buffet: More data & Augmentation
Train for intelligence, not just memory!
Overfitting โ model memorizes training data
Result โ high training accuracy, low test accuracy
Problem โ poor generalization
just exploring......
The Cat in the Box ๐ฑ
Overfitting is when your model memorizes the training "box" so perfectly that it fails to generalize to any other shape
Training: 100% (I fits!)
Testing: 0% (Wrong shape!)
Don't let your models lose the big picture by obsessing over noise
#MachineLearning
Machine Learning Essentials
Overfitting is when a model mistakes random noise for a real signal
Training accuracy looks perfect, but testing accuracy crashes on new data
It happens because the model memorizes specific details instead of learning the general pattern
#AI
Experiment: Photo โ Anime using AnimeGANv2.
Things I expected to do:
Run the model
Things I actually did:
1. Fix TensorFlow 1.x code
2. Replace deprecated APIs
3. Read 200 StackOverflow posts
Tried all these without AI
Worth it though ๐
Original vs Anime result ๐
#AI
Machine Learning โ Day 1
Concept: Train/Test Split
We divide data into:
โข Training set โ model learns patterns
โข Testing set โ model is evaluated
This prevents overfitting.
#MachineLearning#AI
Why is Swin Transformer becoming a strong backbone for computer vision?
Because it solves the biggest problem of Vision Transformers scalability.
Key ideas:
โข Shifted window attention
โข Hierarchical feature maps
โข Lower computation cost