This shortening feature of integration which will

• No School
• AA 1
• 24

This preview shows page 23 - 24 out of 24 pages.

this shortening feature of integration, which will prove wrong on a FWD test set. Similar problems will happen on a FWD trained model with a BWD test set. Since IBP keeps average expression lengths unchanged, BWD and FWD -trained models will generalize better to IBP test sets (and be more accurate on FWD -trained models, since their input length distributions are closer). Figure 2: Distribution of input and output lengths for different integration datasets. The FWD generator produces short problems with long solutions. Conversely, the BWD generator creates long problems, with short solutions. The IBP approach stands in the middle, and generates short problems with short solutions. This suggests that what looks at first glance like a generalization problem (bad accuracy of BWD - trained models on FWD generated sets, and the converse) is in fact a consequence of data generation. BWD and FWD methods generate training sets with specific properties, that our model will learn. But this can be addressed by adding IBP or FWD data to the BWD dataset, as shown in the two last lines of Table 6. In practice, a better approach could be implemented with self-supervised learning, where new training examples are generated by the model itself. 23
Functions and their primitives generated with the forward approach ( FWD ) cos -1 ( x ) x cos -1 ( x ) - p 1 - x 2 x (2 x + cos (2 x )) 2 x 3 3 + x sin (2 x ) 2 + cos (2 x ) 4 x ( x + 4) x + 2 x 2 2 + 2 x - 4 log ( x + 2) cos (2 x ) sin ( x ) log (cos ( x ) - 1) 2 - log (cos ( x ) + 1) 2 + 2 cos ( x ) 3 x 2 sinh -1 (2 x ) x 3 sinh -1 (2 x ) - x 2 4 x 2 + 1 6 + 4 x 2 + 1 12 x 3 log ( x 2 ) 4 x 4 log ( x 2 ) 4 4 - x 4 log ( x 2 ) 3 2 + 3 x 4 log ( x 2 ) 2 4 - 3 x 4 log ( x 2 ) 4 + 3 x 4 8 Functions and their primitives generated with the backward approach ( BWD ) cos ( x ) + tan 2 ( x ) + 2 x + sin ( x ) + tan ( x ) 1 x 2 x - 1 x + 1 x - 1 x + 1 x 2 x cos 2 ( x ) + tan ( x ) tan ( x ) x tan 2 ( x ) x tan e x x + ( x - 1) e x cos 2 ( e x x ) x x tan e x x 1 + 1 log (log ( x )) - 1 log ( x ) log (log ( x )) 2 x + x log (log ( x )) - 2 x 2 sin ( x 2 ) tan ( x ) + x ( tan 2 ( x ) + 1 ) cos ( x 2 ) + cos ( x 2 ) tan ( x ) x cos ( x 2 ) tan ( x ) Functions and their primitives generated with the integration by parts approach ( IBP ) x ( x + log ( x )) x 2 (4 x + 6 log ( x ) - 3) 12 x ( x + 3) 2 - x + ( x + 3) log ( x + 3) x + 3 x + 2 cos 2 ( x ) x + 2 tan ( x ) + log (cos ( x )) x (2 x + 5) (3 x + 2 log ( x ) + 1) x 2 ( 27 x 2 + 24 x log ( x ) + 94 x + 90 log ( x ) ) 18 x - 2 x sin 2 ( x ) + 1 tan ( x ) log ( x ) sin ( x ) x log ( x ) + tan ( x ) sin ( x ) tan ( x ) x 3 sinh ( x ) x 3 cosh ( x ) - 3 x 2 sinh ( x ) + 6 x cosh ( x ) - 6 sinh ( x ) Table 9: Examples of functions with their integrals, generated by our FWD , BWD and IBP approaches. We observe that the FWD and IBP approaches tend to generate short functions, with long integrals, while the BWD approach generates short functions with long derivatives. 24

You've reached the end of your free preview.

Want to read all 24 pages?

• Fall '19

What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern