Cool work done by @sladeka.bsky.social and together with @arnosolin.bsky.social!
Unfortunately, @sladeka.bsky.social has fallen ill and cannot attend @auai.org in person. Please join the virtual poster session at: www.auai.org/uai2025/gath... if you are interested in hearing more about #BitVI.
Entropy of BitVI for varying complexity of the target distribution.
Bayesian neural networks with varying number of numerical precision.
Moreover, #BitVI can help identify the numerical precision required to represent a target density, a crucial task in the quantisation of neural networks and when working with low-precision regimes.
Illustration of fixed-point numbers
Binary decision tree induced by BitVI for fixed-point representations.
Resulting circuit model used in BitVI with fixed-point representations.
In #BitVI, we exploit the fixed-point representations of numbers and a tractable variational approximation based on #circuits. Thus, enabling efficient ELBO computation and control of the numerical precision. Additionally, BitVI is easily extendable to other number systems, such as floating-point.
BitVI on 1D Gaussian mixture models.
Remember that computers use bitstrings to represent numbers? We exploit this in our recent @auai.org paper and introduce #BitVI.
#BitVI directly learns an approximation in the space of bitstring representations, thus, capturing complex distributions under varying numerical precision regimes.
