valensstealth · stanford-advised
a perception company

Vision, before learning.

We give machines the geometry of sight, so they don't have to relearn the world from scratch.

scroll
01 / what we believe

TheThestructurestructureofofvisionvisionisisnotnotaahyperparameter.hyperparameter.ItItisisaafactfactandandweweencodeencodeititdirectly.directly.

02 / what it feels like

A robot that walks into a new room and already knows how light, edges, and motion are shaped.

03 / benchmarks

Under noise, the gap shows up.

3.1 / accuracy
94%
ours
94%
cnn
69%

Accuracy under noise. Standard CNN drops to 69%. Same data, same optimizer.

3.2 / convergence
5×

Faster to 70% action-recognition accuracy on video. 79 frozen parameters.

3.3 / parameters
79
79784,080

Frozen parameters in our representation, vs 784,080 for the discretized equivalent.

3.4 / scaling
O(n)

Cost in input dimension. Previous approaches scale combinatorially. Linear, not polynomial.

04 / why this works

Most networks waste capacity rediscovering geometry that mathematics has already proven.

Small natural image patches concentrate on a Klein bottle — a result proven by Gunnar Carlssonat Stanford. We don't train a network to rediscover that surface. We replace its first layer with a topological lift that already lives on it.

The same construction extends to video through the tangent bundle — the Klein bottle moving through frames. Cost scales linearly in input size, not polynomially. The downstream network has less to learn, and is stable under noise it has never seen.

The place this matters most is robotic perception: data is scarce, environments are noisy, and every new room is a distribution shift.

refCarlsson et al. — On the local behavior of spaces of natural images
05 / who

Built by mathematicians, grounded in topology.

founders
William Ekberg
ceo
M.Sc. Computational Science, Stanford. Transformer optimizer and RL research. Previously Vanguard.
Albin Liljefors
cto
Visiting Researcher, Stanford. Topological deep learning with Gunnar Carlsson. M.Sc. Engineering Physics, Uppsala.
advisors
Gunnar Carlsson
advisor
Professor Emeritus, Stanford. Founder, Ayasdi. Pioneer of topological data analysis.

If the structure is already in the data, the model shouldn't have to relearn it.