Google Office Neural Architecture Device

• 

  Error

• 

  Accuracy after 10 Epochs

Inspiration

Machine learning is the Next Big Thing™, but even they remain unapproachable to the broader public. With Google Office Neural Architecture Device™ (GONAD for short), we bring neural networks to the web.

What it does

GONAD is a neural network classifier that classifies the classic Iris species dataset with a 95% accuracy. And it was written, trained, and validated all a Google spreadsheet document.

How we built it

After intensive research on cloud computing solutions, we decided Google Sheets was the most resource-effective solution. We used Google Sheet formulas to write a backpropagation algorithm that learns the Iris species data set in to >90% accuracy in fewer than 10 epochs.

Challenges we ran into

Our original architecture (1 hidden layer with 3 nodes that had a sigmoid activation function with a sigmoid output) was ineffective, and would not converge. After running some tests, we decided to restart with a softmax activation function and a cross entropy loss function.

Accomplishments that we're proud of

We built a neural network. In Google Sheets.

What we learned

Google Sheets is incapable of loading the MNIST dataset. Don't even try. You'll regret it.

What's next for GONAD

We are currently seeking investors in our new age platform. Remember: it's not a product, it's a movement.

F.A.Q

Why

Why not?

This is stupid

1. That's not a question. 2. Your face is stupid.

What about overfitting?

Totally valid concern. We have yet to implement dropout.

What about underfitting?

Invalid concern. We have a 95% accuracy rate.

Where'd you get the data?

Here.

These aren't frequently asked! You're just making them up!

Yes. Yes, I am.

Built With

• google-sheets