Timezone: »

Stable and expressive recurrent vision models
Drew Linsley · Alekh Karkada Ashok · Lakshmi Narasimhan Govindarajan · Rex Liu · Thomas Serre

Thu Dec 10 07:00 AM -- 07:10 AM (PST) @ Orals & Spotlights: Neuroscience

Primate vision depends on recurrent processing for reliable perception. A growing body of literature also suggests that recurrent connections improve the learning efficiency and generalization of vision models on classic computer vision challenges. Why then, are current large-scale challenges dominated by feedforward networks? We posit that the effectiveness of recurrent vision models is bottlenecked by the standard algorithm used for training them, "back-propagation through time" (BPTT), which has O(N) memory-complexity for training an N step model. Thus, recurrent vision model design is bounded by memory constraints, forcing a choice between rivaling the enormous capacity of leading feedforward models or trying to compensate for this deficit through granular and complex dynamics. Here, we develop a new learning algorithm, "contractor recurrent back-propagation" (C-RBP), which alleviates these issues by achieving constant O(1) memory-complexity with steps of recurrent processing. We demonstrate that recurrent vision models trained with C-RBP can detect long-range spatial dependencies in a synthetic contour tracing task that BPTT-trained models cannot. We further show that recurrent vision models trained with C-RBP to solve the large-scale Panoptic Segmentation MS-COCO challenge outperform the leading feedforward approach, with fewer free parameters. C-RBP is a general-purpose learning algorithm for any application that can benefit from expansive recurrent dynamics. Code and data are available at https://github.com/c-rbp.

Author Information

Drew Linsley (Brown University)

We need artificial vision to create intelligent machines that can reason about the world, but existing artificial vision systems cannot solve many of the visual challenges that we encounter and routinely solve in our daily lives. I look to biological vision to inspire new solutions to challenges faced by artificial vision. I do this by testing complementary hypotheses that connect computational theory with systems- and cognitive-neuroscience level experimental research: - Computational challenges for artificial vision can be identified through systematic comparisons with biological vision, and solved with algorithms inspired by those of biological vision. - Improved algorithms for artificial vision will lead to better methods for gleaning insight from large-scale experimental data, and better models for understanding the relationship between neural computation and perception.

Alekh Karkada Ashok (Brown University)
Lakshmi Narasimhan Govindarajan (Brown University)
Rex Liu (Brown University)
Thomas Serre (Brown University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors