Thanks for sharing, I am working with a hospital to help them take stock of their “AI Readiness”. Do you know of the venues (conference/journals) that are good homes for such work?

Edit: it is *Grassmannian not “Gradsmannian”!

👏The glorious team behind SubTrack your Grad: Sahar Rajabi and Nayeema Nonta #UWaterloo

🔥If you are a PI and want to collaborate, or a student who wants to join my excellent team at #CriticalML drop me a line. We are leading the next wave of innovation at #UWaterloo and are showcasing what inclusive excellence looks like in #AI #ML.

🌱 Being at the forefront of this line of work, I can see phenomenal opportunities for the #foundation #models to become efficient and accessible. Funding such works is critical if we want to be competitive, want to #democratize this transformative technology, and address its #climate impacts!

🎯Just SubTrack your Grad: Turns out that tracking the gradient subspace can reduce wall times by a whopping 65% (15-20 % on a average). We featured our work across workshops at #neurips 2024 with tremendous response and insightful discussions.

🤖While recent works tackle the memory efficiency at the expense of time (GaLore), or time efficiency at the expense of performance (BAdam), our work which tracks the gradient subspace using the Gradsmannian geometry to get you BOTH memory AND time efficiency while being performant!

💜 If you are an #ally or a mentor, please give this signal to the folks around you by advocating for them and listening to their experience. You may catalyze a change in the reward structure, giving them the energy to keep going, perhaps right when they were about to give-up!

🔥You march on and be there for yourself, when you do a tremendous job (who are we kidding, it is pretty much multiple times a day) — tell yourself how amazing that is, and of course bring together allies and support each other. Also, listen to some good rap to get the energy to keep fighting!

👀 So, what do you do?

🎮 In fact, you may get a negative reward when you try to break barriers as it goes against the existing norms, which ultimately may lead you to give-up. This is why it is hard to retain these groups — they are forced to play the game at the hardest setting!

🤖 In a more #RL way of thinking: making decisions based on external validation (rewards) just maintains the status quo. Especially, for minorities & folks with intersectional identities these rewards are very uncertain, and YOU MAY NOT receive this signal to tell you — “Hey, you are great!”

🏆 I remember feeling something similar. It’s easy to fall into the trap of thinking that external accolades are the only way to measure your worth.

🔥Building community is essential to our growth and resilience, and am so grateful for this bunch!

#waterloo #engineering #UWaterloo

💜 We share so much more than we think, and I am always amazed and how much I learn. There are always great ideas and fun times!

🌳 What started as a casual hangout a couple of months ago is now big enough that we need to be planning ahead. And I am happy to channelize my #WiML (Women in Machine Learning) energy to bring people together!