Best Student Paper Award for work in universal training of neural networks

The paper, “Universal Training of Neural Networks to Achieve Bayes Optimal Classification Accuracy,” was selected as a Best Student Paper Award by the 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing (2025 ICASSP) Awards committee. 

First author and recipient of the award is Mohammadreza Tavasoli Naeini (M.S. ECE 2020), a former student who had worked with Prof. Alfred Hero during his time at Michigan. 

“The paper develops, for the first time, a tractable mathematical method for training a neural network to directly optimize its prediction accuracy on the training data,” explains Hero, a co-author on the paper.

“The standard way that neural networks are trained attempts to optimize prediction accuracy by minimizing some heuristically motivated loss function, e.g. relative-entropy and its variants. Our method uses a novel information theoretic representation and bound on the true classification accuracy (average of the false positive and false negative classification errors) to improve the neural network performance.”

Additional co-authors on the paper include: Ali Bereyhi, Assistant Professor of Electrical and Computer Engineering at the University of Toronto; Morteza Noshad (Ph.D. CSE 2020 – advised by Hero), senior ML/NLP scientist at Vida health; and Ben Liang, Professor of Electrical and Computer Engineering at the University of Toronto. 

Hero is the John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science, R. Jamison and Betty Williams Professor of Engineering, and Professor of Electrical and Computer Engineering at the University of Michigan.

The award was announced at 2025 ICASSP on Friday, April 11, 2025.