Active inference is a state-of-the-art framework for modelling the brain that explains a wide range of mechanisms such as habit formation, dopaminergic discharge and curiosity. However, recent implementations suffer from an exponential (space and time) complexity class when computing the prior over all the possible policies up to the time horizon. Fountas et al. (2020) used Monte Carlo tree search to address this problem, leading to very good results in two different tasks. Additionally, Champion et al. (2021a) proposed a tree search approach based on structure learning. This was enabled by the development of a variational message passing approach to active inference (Champion et al., 2021b), which enables compositional construction of Bayesian networks for active inference. However, this message passing tree search approach, which we call branching-time active inference (BTAI), has never been tested empirically. In this paper, we present an experimental study of the approach (Champion et al., 2021a) in the context of a maze solving agent. In this context, we show that both improved prior preferences and deeper search help mitigate the vulnerability to local minima. Then, we compare BTAI to standard active inference (AI) on a graph navigation task. We show that for small graphs, both BTAI and AI successfully solve the task. For larger graphs, AI exhibits an exponential (space) complexity class, making the approach intractable. However, BTAI explores the space of policies more efficiently, successfully scaling to larger graphs.

I have always loved the idea of choosing my own path in a game. Moral dilemmas make virtual worlds more interesting. Sometimes they change the outcome and give you a reason to play the game all over again. But as much as I like the idea, I often struggle to take the evil route. I'll replay a game with the intention of being bad, yet I find myself being a Good Samaritan again.

This paper overview of RL even covers the history, a good summary of a different area of studies. RL has a long history relates to statistic, computer science, and neuroscience. RL agent learns via trial and error it gathers the training data on its own. The standard RL model an agent that learns uses dynamic programming and statistic Not yet clear which method is better overall. For each time stamp, the agent receives some env, reward and more over time optimize the amount of reward it gets over one period.

Sysomos rolled out a unified social media marketing and analytics platform yesterday that it says enables marketers to access all the paid, owned and earned data they need to create strategic campaigns, take action in real time and measure the actions through one interface. In effect, it unifies the range of tools Sysomos has developed or acquired over the years into one platform. Individual users, however, can focus on the aspects that matter most to them, whether it's identifying trending topics, measuring impact or using the refined data to tell relevant stories. The platform also incorporates artificial intelligence to "uncover correlations, anomalies and associations by using machine learning to process trillions of data points every second," as a release puts it, and that's the aspect I'm going to focus on. While viewing a couple of short previews of the new platform that Sysomos CEO Peter Heffring sent over last week, I was struck in particular by its ability to detect patterns not only in the words of a social campaign but also in posted images.