Profile

Cover photo
Ran Manor
Attends Ben-Gurion University of the Negev
719,102 views
AboutPostsPhotosYouTubeReviews

Stream

Ran Manor

Shared publicly  - 
 
אני מתחילון, מה הסטטוס שלך? הנה הקוד שלי GTMVCEJ, שיעניק לך ₪15 לנסיעתך הראשונה. http://invite-il.gett.com/code/GTMVCEJ
 ·  Translate
1
Add a comment...

Ran Manor

Discussion  - 
 
tl;dr - how do I check which features the strongest?

I built a neural network that has two inputs, each of them goes through several layers, and after that the final features are concatenated to one feature vector which is classified in the output layer.

I want to see, for a single sample, which input contributed most to the classification outcome. I thought about looking at the norms of the final features of each network, but the problem is that one input produces 4096 features and the other one produces 40 features, so they are not really comparable, as the norm of the largest vector is always larger.

Anyone has an idea about this?
Thanks!
2
1
Eric Battenberg's profile photoChuck Knowledge's profile photoRan Manor's profile photoEugen Funk's profile photo
6 comments
 
Use crossvalidation when ignoring one of the inpu features (leave one out cv). Apply it on features, not data samples as others do
Add a comment...

Ran Manor

Shared publicly  - 
 
Hey guys, it's giveaway time! Welcome to the Sunday Giveaway, the place where we giveaway a new Android phone or tablet each and every Sunday! A big congratulations to last week’s winners of the BLU Pure XL Giveaway! Sinisa V. (Croatia), Vassilios B. (Greece), and Peter H. (Canada), enjoy your new smartphones! This week we are giving away the LG V10! With its larger display, fingerprint scanner, great design, durable build quality...
1
Add a comment...

Ran Manor

Shared publicly  - 
 
Welcome to the Sunday Giveaway, the place where we giveaway a new Android phone or tablet each and every Sunday. A big congratulations to last week’s winner of the Nexus 5X giveaway: Perver K. (Turkey). This week we are giving away the hotly anticipated Nexus 6P. Launching later this month on the Google Store and retailers from around the world, the Nexus 6P is the pinnacle of the Android experience. The stock Marshmallow software shines...
1
Add a comment...

Ran Manor

Discussion  - 
 
tl;dr - noisy labels in neural networks

I have some data I'm classifying using a neural network. I've noticed that the training labels don't always match the training samples (although most of the time they do match).
Is there anyway to incorporate this in the neural network?
Thanks.
3
1
John Beuving's profile photoBartosz Ludwiczuk's profile photogwern branwen's profile photo
3 comments
 
What do you mean by 'don't match'? As in, this is a classification task with n labels and sometimes a sample is given the wrong label? You don't necessarily need to do anything about this. But what you could do is think about 'active learning' a bit: take the samples which the trained NN makes mistakes on, and check the labels manually & fix them.

Or you could think about your loss function more. You're probably using log loss, but does that loss function really match your needs? Maybe you need more accuracy on particular classes, and mistakes for those classes should be penalized more.

I was thinking about this for multi-class labeling (image tagging): for most such datasets, if an image is tagged with a particular tag, you can be very confident that it matches that tag (a picture with a label like 'lion zebra savannah' almost surely has a lion and a zebra on the savannah in it) but you can't be confident that it doesn't match all the other tags (maybe there's an 'airplane' and a 'cloud' and a 'mountain' in the image, but the human taggers didn't bother with those). So you can use an imbalanced loss function: large penalty on failing to assign high probabilities to provided tags lion.zebra/savannah, but only a small penalty to predicting non-provided-tags airplane/cloud/mountain.
Add a comment...

Ran Manor

Shared publicly  - 
 
Welcome to the Sunday Giveaway, the place where we giveaway a new Android phone or tablet each and every Sunday! A big congratulations to last week’s winners of the Blu Pure XL 3 Phone giveaway : Wayne J. (Canada), Aviral G (India), Rocky S (Canada). This week we are giving away a brand new Nexus 5X The Nexus 5X is offers solid specs, a promising camera, and the Nexus Imprint fingerprint sensor, combined with pure Android and a fast ...
1
Add a comment...

Ran Manor

Shared publicly  - 
 
Google releases its latest machine learning system.
1
Add a comment...

Ran Manor

Shared publicly  - 
 
Use my Uber promo code, erpac, and get ₪30 off your first Uber ride. Redeem it at https://www.uber.com/invite/erpac
A gift from me to you! Claim your first Uber ride free, up to 30 ₪. You'll never need a taxi again. Available on iPhone and Android.
1
Add a comment...

Ran Manor

Shared publicly  - 
 
Welcome to the Sunday Giveaway, the place where we giveaway a new Android phone or tablet each and every Sunday. A big congratulations to last week’s winner of the Oppo R7 giveaway: Nurul S. from Malaysia. This week we are giving away the new OnePlus 2! OnePlus took the wraps off of its latest 'flagship killer' a couple of weeks ago, and the latest iteration continues to offer what we loved about the original, with some refinements and...
1
Add a comment...

Ran Manor

Shared publicly  - 
 
Use my Uber promo code, erpac, and get ₪30 off your first Uber ride. Redeem it at https://www.uber.com/invite/erpac
A gift from me to you! Claim your first Uber ride free, up to 30 ₪. You'll never need a taxi again. Available on iPhone and Android.
1
Add a comment...
Work
Occupation
PhD Student
Skills
drums, guitar and machine learning.
Basic Information
Gender
Male
Story
Tagline
Piled higher and deeper
Introduction
PhD student with interest in machine learning and deep learning.
Education
  • Ben-Gurion University of the Negev
    Electrical & Computer Engineering, 2006 - present
Links
YouTube
Other profiles