SGD with Momentum Explained in Detail with Animations | Optimizers in Deep Learning Part 2

[@ashishmalhotra2230]

your ability to breakdown every concept in its simplest form is amazing! Thank you for making these videos public.

[@ashwnicoer]

No one can explain it better than you sir ! Incredible !

[@jagritikumari1273]

Before a year ago, I heard the term contour plot, but I didn't understand it. My prof has tried to make me know but I can't. My classmates were like you are wasting our time. I felt so dumb at that time. I thought it was my problem that I couldn't understand it. But you are a real guru to me. You have made it very easy for everyone. You know actually where the student gets doubt. I felt so confident after watching your videos. Thank you so much sir for uploading.

[@tanmaygupta8288]

Well said

[@sanjaisrao484]

Thank you for all your efforts sir, You are amazing, Keep continuing the teaching, Please bless me for my future. Thank you

[@3daudiomusic540]

SGD with Momentum is one of the optimizers which is used to improve the performance of the neural network. Let's take an example and understand the intuition behind the optimizer suppose we have a ball which is sliding from the start of the slope as it goes the speed of the bowl is increased over time. If we have one point A and we want to reach point B and we don't know in which direction to move but we ask for the 4 points which have already reached point B. If all 4 points are pointing you in the same direction then the confidence of the A is more and it goes in the direction pointed very fast. This is the main concept behind the SGD with Momentum.

[@himesh5869]

Although, I'm not watching it now because I'm learning ML. But this comment is to say you thank you to keep going. Bohot views aega kuch saalo mein. Take as much time you can (not just 100 days) taaki next competitor(jo bhi baad mein hindi me deep learning video banaye) ko koi dekhe bhi na and you maintain the hegemony.

[@soyam7]

16:20, i literally cried after watching your effort 🥺

[@ThePri1707]

Your focus to build intuitions within your students is what sets you apart! Thanks for the videos.

[@NabidAlam360]

Best playlist ever! I have never seen anyone teaching topics like you! Thank you so much!

[@74kumarrohit]

It seems Nature finally want me to understand ML. I found this Channel. Great work with very good voice and very good teaching technique

[@laxmiagarwal3285]

Very nice explanation This 100 days of deep learning videos has developed my interest in deep learning Thank you so much sir 😊

[@abhimishra_02]

Watching now and it is golden. The content can never be old.

[@vikeshdas3909]

thank you sir to continue the Deep Learning series

[@KapilSharma56419]

World's best explanation

[@ayyamguari6134]

Last explanation diagram was just so good.

[@sameeryadav1130]

Sir ,Thankyou so much please do not stop teaching us. And one request kindly make video on imbalance class .(it seems our 100 days Ml playlist need some more videos )

[@purubhatnagar483]

awesome man, this shows how much dedication you have given to this

[@sameerabanu3115]

Amazing lecture, thank you so much, your efforts are seen sir

[@siddhantsingh9736]

so fortunate that i find your channel : )

[@sahilwithmathphysics8419]

Thank you sir

[@LMessii10]

This is an incredible explanation, the best one out there. It quite baffales me how can somebody explain something in such great details and with such lucidity. I am indebted to you sir ! Brilliance

[@mr.deep.]

Sir thanks🙏🙏🙏

[@rafibasha4145]

Thank you bro

[@ankitchavhan]

Thanks for the insightful video. Visualization helped in understanding the whole concept.

[@vijaypalmanit]

Superb, better explanation and intuition than Andrew NG and Andrez Karpathy

[@teenagepanda8972]

Thank you sir 💯💯

[@vaibhav1468]

Nitish sir, i request you to please complete interview question series. Companies have started visiting campus for placement.

[@rockykumarverma980]

Thank you so much sir🙏🙏🙏

[@amritamukherjee4822]

Thank you sir.

[@thecarradioandarongtong4631]

Really Really Good work done bro. Keep it up ! Indian Channels are generally very shallow, this was not !

[@photogyulai]

The most myterious chanel on youtube. English title, english greeting. English comments, every 20-th sentece is in englist, but everythig else is hindi.... glitch in the matrix

[@raj4624]

God level explanation

[@pavangoyal6840]

Amazing...

[@AbhijeetKumar-cj4fe]

best of best sir

[@rajeshsingh-ds1gt]

Thanks

[@sandipansarkar9211]

finished watching

[@pavantripathi1890]

Thanks for the wonderful explanation!

[@ParthivShah]

Thank You Sir.

[@paragbharadia2895]

btw if this caught your attention, can you tell: where can we play the animation you played at: 37:00 the gd-variant.html too?

[@97m_it]

Hi, Your videos are really great and it's always satisfying to watch your videos and once started I can' t hold myself to keep going. Um, I wanted to try out that last part, I looked for code but coldn't find. Can you provide me the link or anything to try out that.

[@barunkaushik7015]

That is just amazing

[@narendraparmar1631]

Very well explained , Thanks

[@nikitasuryavanshi63]

Very nice...

[@Ajeet-Yadav-IIITD]

Nitish, at 26:00 shouldn't Vt be Vt = beta*Vt-1 + (1-beta)*dw ?

[@vanshshah6418]

Exactly He missed it.

[@JourneytoWINover_PORN]

i was also confused before but that is actually right you can verify it from 29:00

[@ML.ArtificialIntelligence]

Yes, exactly he missed it...

[@Ramapatel12]

🔥🔥🔥🔥🤩🤩

[@muhammadfahad9946]

Thanks Sirg Sir kindly paste link of WEB visualization of optimizers....

[@rafibasha4145]

Please share the video about convex and non convex which you mentioned

[@campusx-official]

kzbin.info/www/bejne/iom5pZWdlqqso9U

[@rafibasha4145]

@@campusx-official thanks bhai

[@shaktijain8560]

Amazing ❤

[@rajbir_singh0517]

bahut acha laga sir

[@Sara-fp1zw]

you are best

[@InstantMOVIES7]

Nice na

[@learnfromIITguy]

exellent

[@a.syndeed]

Great video. But the formula is: v_current = - beta*v_prev + learning_rate*grad_wrt_weights. You didn't include the -ve sign in the beginning. The momentum is added to the weights, not subtracted. So the weight_update formula goes: w_next = w_current + beta*v_prev - learning_rate*grad_wrt_weights

[@abdulrauff1150]

you are amazing. I got every bit from this lecture.❤

[@JourneytoWINover_PORN]

to kya krein

[@rajeshsingh-ds1gt]

Sir aap jo paid batch lete the us ka recording mill sak ta hai kya please sir reply muhe deep learning jaldi khatam kar ne ka hai sir please please please reply 🙇🙇🙇🙇🙇🙇🙇🙇🙇🙇🙇🙇🙇

[@JourneytoWINover_PORN]

paid batch ke paise de pehle

[@akashdas8977]

But sir my question is that irrespective of were it should go to the global mini but why it has convert it to the local mini na at the first iteration?

[@ShubhamVishwakarma-bz5sz]

Sir can you please link the code for the web based comparator of optimizers

[@RushikeshPatel-s4k]

can anyone explain why there two different formulas for GD with momentum in this lecture and andrew ng's lectures.

[@miscellaneoushub1956]

sir, can you please provide references books/research paper etc for the concept like saddle point, high curvature low radius.

[@kindaeasy9797]

what a treat !!

[@NKONDURURAGHUSURYA]

what is Vt-1? Is it the previous gradient or previous step(gradient multiply with learning rate)?

[@reetanshukumarpandey1909]

Previous velocity , nothing to do with learning rate directly,but indeed to find vt-1 ,you need vt-2 ,hence leaning rate for vt-2 , so they are all related in a chain...

[@JourneytoWINover_PORN]

ig it is the velocity of slope of just previous step

[@joyal7701]

nice

[@JANARDHANREDDYILLURU]

sir could you please share the visualization tool used at end of the video

[@alroygama6166]

Sir how can i access this tool?

[@debanjanbhattacharjee9477]

why in high curvature traversing is difficult?

[@ali75988]

just a small change in x and loss would overshoot.

[@farheensyed4529]

May i know what is the tool name of this visualization for optimizers ?

[@MayankKashyap-k8j]

code kaha hai?

[@yashjain6372]

best

[@ajitkumarpatel2048]

🙏

[@angadshagun8339]

Sir please send this visualizer tool file

[@vanshshah6418]

You Made a mistake. Actually vt=B*(vt-1)+(1-B)*gradient. You Missed The term (1-B) to multiply it with loss. Thanks though

[@campusx-official]

It's optional actually. A lot of places you would see people omitting it

[@JourneytoWINover_PORN]

is (1-beta) =n ,there?

[@raadonyt]

✨✨✨✨

[@alastormoody1282]

SIR

[@MohammedArsath-p8t]

you explained well but in hindi. can u explain in english

[@jyothiyadav2595]

Amazing ❤