arXiv subjects

Aditya Paliwal

Publications and source records attributed to Aditya Paliwal.

3 recordsLinked to original sources

arXiv2020

Reinforced Genetic Algorithm Learning for Optimizing Computation Graphs

We present a deep reinforcement learning approach to minimizing the execution cost of neural network computation graphs in an optimizing compiler. Unlike earlier learning-based works that require training the optimizer on the same graph to be optimized, we propose a learning approach that trains an optimizer offline and then generalizes to previously unseen graphs without further training. This allows our approach to produce high-quality execution decisions on real-world TensorFlow graphs in seconds instead of hours. We consider two optimization tasks for computation graphs: minimizing running time and peak memory usage. In comparison to an extensive set of baselines, our approach achieves significant improvements over classical and other learning-based methods on these two tasks.

cs.LG↗

arXiv2019

Graph Representations for Higher-Order Logic and Theorem Proving

This paper presents the first use of graph neural networks (GNNs) for higher-order proof search and demonstrates that GNNs can improve upon state-of-the-art results in this domain. Interactive, higher-order theorem provers allow for the formalization of most mathematical theories and have been shown to pose a significant challenge for deep learning. Higher-order logic is highly expressive and, even though it is well-structured with a clearly defined grammar and semantics, there still remains no well-established method to convert formulas into graph-based representations. In this paper, we consider several graphical representations of higher-order logic and evaluate them against the HOList benchmark for higher-order theorem proving.

cs.LG↗

arXiv2018

Distributed Approximation Algorithms for the Combinatorial Motion Planning Problem

We present a new $4$-approximation algorithm for the Combinatorial Motion Planning problem which runs in $\mathcal{O}(n^2α(n^2,n))$ time, where $α$ is the functional inverse of the Ackermann function, and a fully distributed version for the same in asynchronous message passing systems, which runs in $\mathcal{O}(n\log_2n)$ time with a message complexity of $\mathcal{O}(n^2)$. This also includes the first fully distributed algorithm in asynchronous message passing systems to perform "shortcut" operations on paths, a procedure which is important in approximation algorithms for the vehicle routing problem and its variants. We also show that our algorithm gives feasible solutions to the $k$-TSP problem with an approximation factor of $2$ in both centralized and distributed environments. The broad idea of the algorithm is to distribute the set of vertices into two subsets and construct paths for each salesman over each of the two subsets. Finally we combine these pairwise disjoint paths for each salesman to obtain a set of paths that span the entire graph. This is similar to the algorithm by Yadlapalli et. al. \cite{3.66} but differs in respect to the fact that it does not require us to use minimum cost matching as a subroutine, and hence can be easily distributed.

cs.DM↗