🧵 Tropical Attention --> Softmax is out, Tropical max-plus is in 🦾.1/ 🔥Ever experinced softmax attention fade as sequences grow?.That blur is why many attention mechanisms stumble on algorithmic and reasoning tasks. Well, we have a Tropical solution 🌴

R2: timely study, extremely bold and ambitious, mentioning NO weaknesses at all, only one question of future directions, then 2: reject….WTH!!!.#NeurIPS2025.

RT @ai4mathworkshop: It's happening today!.📍Location: West Ballroom C, Vancouver Convention Center.⌚️Time: 8:30 am - 6:00 pm.🎥 Livestream:….

The AI for Math workshop papers are out, the studies were of really high quality 🦾.Come to our #AI4Math workshop tomorrow. It is gonna be epic:.

RT @getjonwithit: The desiccated "Theorem, Lemma, Proof, Corollary,. " presentational style is staggeringly counterproductive, if one's ob….

Excited to present our recent work on Trooical Algebraic Geometry, Transformers and Neural Algorithmic Reasoning.

Here is the source code for the Tropical Attention kernel:.

It is still surprising to me how such a deep structure, Grothendieck-Riemann-Roch theorem, arises from merely a book-keeping dictionary, aka Divisors.

RT @gm8xx8: Tropical Attention: Neural Algorithmic Reasoning for Combinatorial Algorithms. Tropical Attention uses a max-plus semiring mech….

7/ Our message ✍️.Better reasoning might come not from bigger models, but from choosing the right algebra/geometry 🌴. @PetarV_93 @fedzbar @lorgiusti @DobrikG @AldenHung @_onionesque @miniapeur .#TropicalGeometry #NeuralAlgorithmicReasoning #AI4Math @fly51fly @Jose_A_Alonso.

6/ We also show that each Tropical attention head can function as a tropical gate in a tropical circuit, simulating any max-plus circuit.

5/ We benchmarked on 11 canonical combinatorial tasks. Tropical attention beat vanilla & adaptive softmax attention on all three OOD axes, Length, value and Adversarial attack generalization:

4/ Tropical Attention runs each head natively in max-plus. Result: .Strong OOD length generalization with sharp attention maps even in several algorithmic tasks, including the notorious Quickselect algorithm (Another settlement for the challenge identified by @michael_galkin)

3/ In the Tropical (max + ) geometry, “addition” is max, “multiplication” is +. Many algorithms already live here, carving exact polyhedral decision boundaries --> so why force them through exponential probabilities?.Let's ditch softmax, embrace the tropical semiring 🤯🍹.

2/ We introduce Tropical Attention -- the first Neural Algorithmic reasoner that operates in the Tropical semiring, achieving SOTA OOD performance on executing several combinatorial algorithms,

RT @ai4mathworkshop: Glad to announce the ICML 2025 Challenge on Automated Math Reasoning and Extensions! 🌟🧮⚛️ . Track 1: File-level Automa….

One of those big moments incoming… I’ll be sharing the full results very soon. Stay tuned!

RT @kagglingdieter: If you dream of getting better at anything — data science, Kaggle, or life — remember this:.Talent helps. Knowledge hel….

If one tries to observe below the Planck time, the very act of observation using any apparatus exeeds its own Schwazschild radius and triggers gravitational collapse and a black hole.

RT @tejotaefe: This is best thing I read recently. It shows how Riemann's moduli problem permeated the mathematical zeitgeist for almost a….