Michal Wilinski
@inverse_hessian
Followers
181
Following
1K
Media
10
Statuses
221
member of technical staff @ stealth & incoming phd student @SCSatCMU · bsc from @PUT_Poznan
Poland
Joined November 2016
RT @OWW: Beyond Constant Parameters: Hyper Prediction Models and HyperMPC.
arxiv.org
Model Predictive Control (MPC) is among the most widely adopted and reliable methods for robot control, relying critically on an accurate dynamics model. However, existing dynamics models used in...
0
1
0
RT @gm8xx8: μ-Parametrization for Mixture of Experts. µP-MoE extends µParameterization to MoE Transformers, enabling zero-shot learning rat….
0
15
0
RT @TimDarcet: hey we heard you liked dinov2 so we got you more of the same shit. dinov3 is like dinov2 in the sense that it's much better….
0
11
0
RT @cloneofsimo: Deepseek is great but @kuba_krj did this first in feb 2024.With explicit scaling law. Which was on top of @_aidan_clark_….
0
10
0
This week I'm presenting some works at @kddconf in Toronto 🇨🇦. Let’s connect if you’re interested in privacy/gradient inversion attacks in federated learning, counterfactual explanations, or fairness and xai!. Here’s where you can find me:.
0
2
6
RT @kellerjordan0: New NanoGPT training speed record: 3.28 FineWeb val loss in 2.863 minutes on 8xH100. New record-holder: @.ClassicLarry o….
0
33
0
RT @marekkraft: We are opening a second @esa project at @PUT_Poznan - this time we're looking for student interns! If you're into space rob….
0
2
0
RT @willccbb: imagine if gas stations didn't tell you how many gallons you were getting because car mileage was a trade secret and the gas….
0
179
0
RT @mervenoyann: spent few hours integrating this to TRL for online methods 🤝🏻 the code itself isn't much but testing took time 🥲. works wh….
0
6
0
RT @huybery: We’ve updated Qwen3 and made excellent progress. The non‑reasoning model now delivers significant improvements across a wide r….
0
62
0
RT @lmthang: Very excited to share that an advanced version of Gemini Deep Think is the first to have achieved gold-medal level in the Inte….
0
233
0
RT @Mihonarium: 🚨 According to a friend, the IMO asked AI companies not to steal the spotlight from kids and to wait a week after the closi….
0
198
0
RT @PiotrRMilos: Today, come and see us at the poster session; East Exhibition Hall. - Joint MoE Scaling Laws (E-2609); tl;dr MoE can be me….
0
5
0
Armed with this understanding, we prune redundant layers from the model. The result? Significantly faster inference and improved efficiency with good performance across various time series tasks!
1
1
1
In the second part of our work, we leverage insights from learned representations to boost TSFM efficiency. We start by assessing redundancy across model layers, examining similarities in their learned representations.
1
0
0
Next, we identify meaningful concept vectors corresponding to these temporal properties, allowing us to actively steer model outputs!
1
0
0