From Tutorial on Active Inference by Oleg Solopchuk

*ise *…possible. How does it connect to active inference and an agent that avoids surprising observations? In fact, **maximizing model evidence is equivalent to minimizing surprise**, which is just a negative log of *p(o)*. If probability is 1 — surprise is 0, probability is 0 — surprise is infinite. Here is surprise as …

From Tutorial on Active Inference by Oleg Solopchuk

…etter our model, the higher will be the probability of the *obse*rved data p(o). It is also called 1*) ‘mod*el evidence’*, since* it quantifies how well is our model pred*icti*…etter our model, the higher will be the probability of the observed data p(o). It is also called 1) ‘model evidence’, since it quantifies how well is our model predicting the real data, 2) ‘marginal likelihood’, beca…

From Tutorial on Active Inference by Oleg Solopchuk

… called 1) ‘mo*del ev*idence’, sin*ce it q*uantifies how well is our model predictin*g th*… called 1) ‘model evidence’, since it quantifies how well is our model predicting the real data, 2) ‘marginal likelihood’, because we marginalize, or sum out, the hidden state s. Compare 2 models below, in which we estima…

From Tutorial on Active Inference by Oleg Solopchuk

s to remain alive, by maintaining its homeostasis. To this end, the agent must ensure that important parameters, like body temperature or blood oxygenation, don’t deviate too much from the norm, i.e. are not surprising. But since it’s only possible to infer these parameters from sensory m…vation of active inference is that the agent wants to remain alive, by maintaining its homeostasis. To this end, the agent must ensure that important parameters, like body temperature or blood oxygenation, don’t deviate too much from the norm, i.e. are not surprising. But since it’s only possible to infer these parameters from sensory measurement…

From Tutorial on Active Inference by Oleg Solopchuk

… parameters, like body temperature or blood oxygenation, don’t deviate too mu… parameters, like body temperature or blood oxygenation, don’t deviate too much from the norm, i.e. are not surprising. But since it’s only possible to infer these parameters from sensory measurements, the agent minimi…

From People Ask Me, What Do You Have Against Deep Learning? by Louis Savain

…have never seen before. This is crucial to survival. A deep learning system would be blind to them. We only remember important high level bits and pieces of the patterns that we see. Most of the low level details are either forgotten or are written over by new experiences.

From Bayesian Nonparametrics by Vadim Smolyakov

…pses (samples from posterior mixture distribution) of the DPMM after 100 Gibbs sampling iterations. The DPMM model initialized with 2 clusters and a concentration parameter alpha of 1, learned the true number of clusters K=5 and concentrated around cluster centers.

From Bayesian Nonparametrics by Vadim Smolyakov

…pi_k. The length of the piece that we break off is determined by the concentration parameter alpha. For alpha=5 (middle) the stick lengths are longer and as a result there are fewer significant mixture weights. For alpha=10 (right) the stick lengths are shorter and therefore we have more significant components.