Daniel Augusto

@spectral.space

102 Followers 398 Following

PhD student at UCL. Working on Gaussian Processes. I love everything mathy, videogames, and language. 🇧🇷 He/Ele/Él Also as @spectraldani@sigmoid.social

Night sky in Brazil showing some stars -- with focus on what I believe to be orion's belt
A wooden model of a tree with pixelated kanji made with beads
a photo of myself on the kanji dictionary's kanji wall showing the character 浦
Photo of the night sky in London with Mars and the two brightest stars of the Gemini constellation. The stars seems to lie on the same line.
A screenshot of our paper, by Tiago da Silva, me, and Diego Mesquita. Abstract: Bayes' rule naturally allows for inference refinement in a streaming fashion, without the need to recompute posteriors from scratch whenever new data arrives. In principle, Bayesian streaming is straightforward: we update our prior with the available data and use the resulting posterior as a prior when processing the next data chunk. In practice, however, this recipe entails i) approximating an intractable posterior at each time step; and ii) encapsulating results appropriately to allow for posterior propagation. For continuous state spaces, variational inference (VI) is particularly convenient due to its scalability and the tractability of variational posteriors, For discrete state spaces, however, state-of-the-art VI results in analytically intractable approximations that are ill-suited for streaming settings. To enable streaming Bayesian inference over discrete parameter spaces, we propose streaming Bayes GFlowNets (abbreviated as SB-GFlowNets) by leveraging the recently proposed GFlowNets -- a powerful class of amortized samplers for discrete compositional objects. Notably, SB-GFlowNet approximates the initial posterior using a standard GFlowNet and subsequently updates it using a tailored procedure that requires only the newly observed data. Our case studies in linear preference learning and phylogenetic inference showcase the effectiveness of SB-GFlowNets in sampling from an unnormalized posterior in a streaming setting. As expected, we also observe that SB-GFlowNets is significantly faster than repeatedly training a GFlowNet from scratch to sample from the full posterior.
A view of Vancouver with a calm waterfront in the foreground, buildings lining the shore, and a forested mountain range in the background topped with patches of snow.
An evening scene showing a calm river reflecting soft lights from nearby street lamps and houses. The foreground features trees framing the view, while in the distance, a small bridge and hillside buildings are visible. The sky above is painted in light blue with delicate, wispy pink clouds, contrasting the darker, shadowed landscape below.
A close up shot of a next train indicator in a London train station. It shows that the next train to London St. Pancras will arrive in 1158 minutes.
A photo of some woods around London. The floor is brown-ish due to fallen leaves while narrow trunk trees are seen all around, these trees have green moss in their trunks and eventually color the upper side of the photo with green.
Photo of the statue of George I atop the spire of St George's Church. It depicts the king pointing down with what I believe are Roman clothes. It's a clear day with almost no clouds.