Goto

Collaborating Authors

 Deep Learning




Supplementary Material: Simultaneous embedding of multiple attractor manifolds in a recurrent neural network using constrained gradient optimization

Neural Information Processing Systems

The dynamics of neural activity are described by a standard rate model. Note that only the third term of Eq. 'th place cell preferred firing position in the's are standard unit vectors spanning an orthonormal basis. To derive Eq. 3 we evaluate the derivative of Energy landscapes were uniformly shifted throughout the manuscript by a constant (Figs. For each network with a different number of total embedded maps, 15 realizations were performed in which the permutations between the spatial maps were chosen independently and at random. Code availability Code is available at public repository https://doi.org/10.5281/zenodo.10016179.







Towards Automated Circuit Discovery for Mechanistic Interpretability

Neural Information Processing Systems

Through considerable effort and intuition, several recent works have reverse-engineered nontrivial behaviors of transformer models. This paper systematizes the mechanistic interpretability process they followed. First, researchers choose a metric and dataset that elicit the desired model behavior.


Expert load matters: operating networks at high accuracy and low manual effort

Neural Information Processing Systems

In human-AI collaboration systems for critical applications, in order to ensure minimal error, users should set an operating point based on model confidence to determine when the decision should be delegated to human experts.