This paper presents a novel optimization for differentiable programming named coarsening optimization. It offers a systematic way to synergize symbolic differentiation and algorithmic differentiation (AD). Through it, the granularity of the computations differentiated by each step in AD can become much larger than a single operation, and hence lead to much reduced runtime computations and data allocations in AD. To circumvent the difficulties that control flow creates to symbolic differentiation in coarsening, this work introduces phi-calculus, a novel method to allow symbolic reasoning and differentiation of computations that involve branches and loops. It further avoids "expression swell" in symbolic differentiation and balance reuse and coarsening through the design of reuse-centric segment of interest identification. Experiments on a collection of real-world applications show that coarsening optimization is effective in speeding up AD, producing several times to two orders of magnitude speedups.

In 2015-16, San Francisco Unified School District (SFUSD) piloted MyCS, a Middle-years CS curriculum, in half of its middle schools. This unexpected launch led naturally to diverging curricular branches: one that evolved within the district, another used by schools with the program already in place. The summer after the pilot, SFUSD's and MyCS's stewards convened for a week of feedback, PD, and planning. This poster highlights the curricular refinements, preliminary assessment results, and institutional changes that came from this curricular divergence and subsequent reconciliation. The data analyzed include teacher- and district-feedback, along with an analysis of student responses from SF's pilot implementation. Though accidental, this experiment suggests that substantial benefit can come from independently co-evolving (branching) and then reconciling (merging) curricula. When merged, those otherwise independent branches create a community both stronger and more invested for all of its stakeholders.