Future Research

We will focus on devising particularly compact, particularly reasonable sets of initial codes with particularly broad practical applicability. It may turn out that the most useful initial languages are not traditional programming languages similar to the FORTH-like one from Section A, but instead based on a handful of primitive instructions for massively parallel cellular automata [73,76,82], or on a few nonlinear operations on matrix-like data structures such as those used in recurrent neural network research [78,42,5]. For example, we could use the principles of OOPS to create a non-gradient-based, near-bias-optimal variant of the successful recurrent network metalearner by [16]. It should also be of interest to study probabilistic Speed Prior-based OOPS variants [58] and to devise applications of OOPS-like methods as components of universal reinforcement learners (Section 5.3). In ongoing work, we are applying OOPS to the problem of optimal trajectory planning for robotics [66] in a realistic physics simulation. This involves the interesting trade-off between comparatively fast program-composing primitives or ``thinking primitives'' and time-consuming ``action primitives'', such as stretch-arm-until-touch-sensor-input (compare Section 4.5).

Back to OOPS main page