waveform-inversion

[thorny perch]

anyone want to collaborate?

[thorny perch]

not sure yet, but seems like a cool competition

[thorny perch]

yeah

[upbeat radish]

definitely 🙂

[thorny perch]

haha what's all this

[upbeat radish]

DOWSING, is an ancient practice using a forked stick, rod, or pendulum to locate underground water or other hidden substances. The "waveforms" in this context refer to the perceived subtle vibrations or energy fields that dowsers believe are emitted by the hidden object and sensed by the dowsing tool. These vibrations, according to dowsers, cause the rod or pendulum to move or bend, indicating the location of the desired substance.

[thorny perch]

ohh ok thank u

[tardy umbra]

Thought u were jk like always

[tardy umbra]

Kindly Share ur approach

[upbeat radish]

There is always Method to my Madness 😉

[zenith glade]

"I have a plan. I have part of a plan. I don't know, 12%" "That is not a plan" "It's barely a concept."
Achieve a minimum viable product (MVP) based on:

• 2d modeling of heterogeneous regions for now
• Model with transformers on the ML side
• Model on physics side with
  • Radial emitter (modeled as radial boundary see below)
  • Point receivers
  • Boundaries (wave in -> reflected wave + transmitted wave)
  • Waves (starting boundary, transmission direction along boundary, transmission properties along boundary (decay, etc))
    Burndown Risk List:
• Heterogeneous is known bad assumption for Waveform Inversion
• ML Transformers are computationally expensive
• Unsure this reduces dimension space ML needs to learn from the training data or...
• May generalize poorly
  Risk Mitigation:
• Handle homogeneous problem with learned propagation waveforms (needs to be constrained or recreates entire problem)
• Verify computational cost early with MVP
• Compare to public SOTA code with small data set (genuinely smaller learning space-> should learn better even with small data sets)
• May leave for later, but purposely develop A/B test sets of significantly different geometries but this may exceed the competition expectations and be unnecessary.

[zenith glade]

(This is mentioned on Kaggle, mentioning here since they have direct links to many of their WFI papers)
https://smileunc.github.io/

[scenic gate]

I am interested in becoming a member of teams which require team members.

Feel free to shoot me a DM.

[sonic gust]

Hello

[dark pecan]

Hello, my team ran into an error. We think the temp.txt file is missing, do you guys know if it's missing and how we can get it?

[hearty sapphire]

Hi. As far as I know, the competition does not have a “temp.txt file” Could you please be more specific? What do you need to use this file for

[dark pecan]

Gotcha let me check back, it was an error when we were running the Github code. Wasn't sure if it was missing. Thanks!

[hearty sapphire]

lmk how that goes 🙂

[stone shuttle]

I also want to collaborate

[stone shuttle]

How do we know when a team can start coding for the competition?

[pliant grail]

you just join and start

[hearty sapphire]

The competition already started. You can register for it on the competition page and start submitting up to 5 times a day

[stable crown]

Do we have a run time limit for the notebooks? Tried to train one of the unet notebooks and it seems to have stop at the 10ish h mark, while still only at 75% of the epochs.
And I still had 19ish hours left of free gpu usage

[drifting smelt]

hii im a first year student and decided to take part in my first kaggle competition. but i feel so lost and stuck. i went through the tutorial but im really not sure what i'm suppose to do next. could anyone please guide me? tysm in advance! i'd really appreciate it 💗

[cosmic creek]

Hi everyone! 👋 I'm new to machine learning and really excited about the Waveform Inversion competition. I'm looking to join a team or collaborate with others who might also be learning or open to helping a beginner grow. If any team has a spot available or if anyone wants to start a new team together, I’d love to connect. Feel free to DM me—thanks! 😊

[vital wadi]

My personal recommendation - if its your first competition, start with something simple like one which has less dataset

This one is 100GBs of dataset and gonna take lot of computation and time

[late pilot]

I'm trying to replicate vel-map generation especially for CurveFault/CurveVel! if you have any idea for parameter range or num of application, i'd love to hear about that!
https://www.kaggle.com/competitions/waveform-inversion/discussion/585539

[cobalt night]

someone plz block these annoying spammers!