This work is both theoretical and empirical. The theoretical part concerns the framework, whereas the empirical part is an analysis of the survey responses, mostly through correlation analysis.

I appreciated the author's responses, and I think the proposed refinements will strengthen the manuscript. As such, I decided to increase my score: 5 - 6. However, I remain lukewarm regarding the actual results shown in the paper. I found the comparisons to be limited (the authors still resist performance comparisons with common approaches such as GPFA or LDS in Figure 1), and the performance quantifications to not be very elucidating (they are focused solely on modest gains in predictive performance whereas the strongest motivation for this method is interpretability). Given that what I find most exciting in this submission is the potential for interpretability, I'm pretty disappointed no effort is done to explore this avenue in the results. To be clear, I agree that it is unreasonable to expect fully featured scientific results in a NeurIPS submission, but I would have liked to at least see this interpretability aspect briefly explored.

The paper is well written, although *very* dense both in terms of mathematical expectation and development, as well as in terms of space. It is *not* an easy read (I suppose unless you are super fluent in LDDMM background). I think the authors could improve this to help this paper reach a broader audience, or perhaps they are not interested in this, I'm not sure. As it stands, it is a *bit* hard to evaluate due to the super condensed and dense nature of it. I believe this is a technical clean contribution with a clear advancement.

This paper considers the variational approach to the inference problem on certain type of temporal graphical models. It defines a convex formulation of the Bethe free energy, resulting in a method with optimal convergence guarantees. The authors derive Expectation-Propagation fixed point equations and gradient-based updates for solving the optimization problem. A toy example is used to illustrate the approach in the context of transportation dynamics and it is shown that the proposed method outperforms sampling, extended Kalman Filtering and a neural network method. In the variational formulation, the optimization problem is generally non-convex, due to the entropy terms.

The paper intends to unify several previously-achieved characteristics of movement primitives (MP) in a single probabilistic framework, while also describing new ways in which this framework allows MPs to be modified or combined. This is accomplished through the representation of trajectories by probability distributions of joint location and velocity. The authors present the foundation for their formulation by describing how standard MP features such as rhythmic and stroke-based movements and temporal modulation are achieved in this new framework. They also discuss how, due to the probabilistic nature of this framework, they can modify position and velocity of a given trajectory through conditioning as well as blend multiple MPs together by multiplying distributions. The authors derive the necessary values for a robotic controller and conclude the paper with experiments on both a real and simulated robotic arm.

Our dependency on AI-based outputs seems to grow every day, both from a business as well as personal perspective. But are we willing to fully trust this output? Are we sure the data fed into these systems is accurate? Are the decision models and algorithms kept up to date? Are humans kept in the loop?

Currently employed neural network architectures have mostly been developed manually by human experts, which is a time-consuming and error-prone process. This is when Neural architecture search, a subset of AutoML, came to the rescue. Neural Architecture Search (NAS) is the process of automating architecture engineering. Here we list top research works in Neural Architecture Search based on their popularity on Github. These works have set new baselines, resulted in new networks and more.

There's no doubt business leaders see artificial intelligence as the way to get more things done around their organizations. A majority of executives in a recent survey, 62%, believe AI will help drive efficiency and competitiveness. The only catch is, to get to this point where machines are picking up the cognitive work previously employed by humans, they need more humans who can build these systems. AI talent is part technological, part business savvy. That's one of the takeaways from an EY study of 800 CEOs and business leaders, who are all excited about AI.

As one (human) study participant put it: "I felt very stressed competing with the robot. In some rounds, I kept seeing the robot's score increasing out of the corner of my eye, which was extremely nerve-racking."

Artificial intelligence (AI) and big data aren't just something quants and scientists can love. CEOs, CFOs and everyone else are also taking a shine to these technology developments as well. They see it as their best hedge against those pesty disruptive competitors that keep shaking up their markets. Which is a shame, because the "big data" side of the equation has been around for more than a decade at this point. The technology exists and is affordable, but many organizations have been slow to adopt and adapt their corporate cultures to embrace data-driven decision-making.