On the ability and limitations of transformers to recognize formal languages

S Bhattamishra, K Ahuja, N Goyal - arxiv preprint arxiv:2009.11264, 2020 - arxiv.org
Transformers have supplanted recurrent models in a large number of NLP tasks. However,
the differences in their abilities to model different syntactic properties remain largely …

Simplicity bias in transformers and their ability to learn sparse boolean functions

S Bhattamishra, A Patel, V Kanade… - arxiv preprint arxiv …, 2022 - arxiv.org
Despite the widespread success of Transformers on NLP tasks, recent works have found
that they struggle to model several formal languages when compared to recurrent models …

Learning deterministic weighted automata with queries and counterexamples

G Weiss, Y Goldberg, E Yahav - Advances in Neural …, 2019 - proceedings.neurips.cc
We present an algorithm for reconstruction of a probabilistic deterministic finite automaton
(PDFA) from a given black-box language model, such as a recurrent neural network (RNN) …

Separation of memory and processing in dual recurrent neural networks

C Oliva, LF Lago-Fernández - International Conference on Artificial Neural …, 2021 - Springer
We explore a neural network architecture that stacks a recurrent layer and a feedforward
layer, both connected to the input. We compare it to a standard recurrent neural network …

Recurrent Neural Networks for Robotic Control of a Human-Scale Bipedal Robot

JA Siekmann - 2020 - ir.library.oregonstate.edu
Dynamic bipedal locomotion is among the most difficult and yet relevant problems in modern
robotics. While a multitude of classical control methods for bipedal locomotion exist, they are …