Editorial

The "Symbolic Computation and Machine Learning" (SCML) forum is an initiative of the Research Institute for Symbolic Computation (RISC) at the Johannes Kepler University (JKU), Austria, pursued in close collaboration with other institutions, in particular with the Journal of Symbolic Computation (JSC).

SCML provides a dedicated publication space for research on the interaction between Symbolic Computation (SC) and Machine Learning (ML). It features a continuous series of virtual workshops with an outlet for refereed proceedings and an opportunity to publish polished versions of accepted papers in regular issues of the JSC; these papers are also collectively available as a "virtual special issue" of the JSC on SCML. Physical instances of a corresponding conference may follow. The details of the submission, reviewing, and publishing procedure of SCML, as well as a list of research topics that are in the scope of SCML, are described in the call for papers.

The core focus of SCML is research that explicitly connects SC and ML as two major approaches towards the goal of Artificial Intelligence (AI). Therefore we only consider submissions that explore the interaction between the two fields - not standalone works on either SC or ML. The interest in the interaction between SC and ML has recently taken speed in both communities. This appears to be a natural consequence of the view that AI is fundamentally about automating the process from a problem specification to methods/algorithms/programs/tools that solve the problem. Towards this goal, over the decades the two distinct AI paradigms of SC and ML have emerged:

  • SC is appropriate for problems that can be specified "in general terms", i.e., by a formal text that describes the desired properties of the output in dependence on the input. For such problems, SC tries to develop a general method (an "algorithm") that operates in a step-by-step process on "symbolic" (formal text) objects which is understandable to humans and yields a provably correct output.
  • ML is appropriate for problems that can be only specified "by examples", i.e., by input/output pairs. For such problems, ML tries to construct an approximative method (the "model") that adequately covers the given examples (the "training set") and is also likely to generalize to new ones, but without guarantee and operating in a way that can be hardly interpreted by humans.

Since AI became an explicit topic in mathematics/computer science (around 1960), in the various "summers" and "winters" of the field, both approaches to AI had their success stories – and also frustration. The success stories, with a few exceptions like computer chess or the formal verification of complex hardware, did not obtain the attention of the broad public. This changed drastically with the advent of the "Large Language Models" (LLMs) in the past couple of years because meaningful processing of natural language text changes the life of basically everybody in an impressive way.

The AI techniques behind LLMs are basically ML techniques. This made many researchers believe that the time of the SC approach is over. However, more and more, it becomes clear that there is a limitation to the ML approach, and the next big step in AI sophistication will need the amalgamation of the two approaches in an appropriate way. This is the conviction that stands behind our initiative. We want to give a forum to research that brings the SC and the ML techniques together for reaching higher and higher levels of AI. We encourage researchers – in particular young researchers - from all over the world to participate in this exciting endeavor. We will need the best minds.