⚡

Terrence Sejnowski

Original nameTerrence Sejnowski

American Computational Neuroscientist, Co-Inventor of the Boltzmann Machine

The Math Behind the Algorithm · Theoretical Founders

The Boltzmann MachineComputational NeuroscienceThe NETtalk Speech Synthesis System

Who they are

Terrence Sejnowski (1947– ) is an American computational neuroscientist who with Geoffrey Hinton in 1985 co-proposed the "Boltzmann machine," an early neural-network model designed drawing on the energy-minimization principle of statistical physics, giving important theoretical inspiration for the later development of several deep-learning architectures. The NETtalk system Sejnowski thereafter developed could learn to convert English text into corresponding speech, and the visualization of the neural network gradually learning pronunciation rules during training left a deep impression on many researchers at academic conferences of the time, one of the earliest vivid demonstrations of the idea that neural networks have the ability to learn. Sejnowski long devoted himself to promoting the cross-fusion of computational neuroscience and AI, holding that understanding the actual working mechanism of the biological brain can provide a continual source of inspiration for designing more effective artificial neural networks.

Primary sourcesAckley, Hinton & Sejnowski, "A Learning Algorithm for Boltzmann Machines" (1985)Sejnowski, The Deep Learning Revolution (2018)

Key stories

A System That "Learned" How to Read English

In the early stage of training, the NETtalk system Sejnowski developed read English words in a slurred, near-noise pronunciation, but as training continued the system’s output pronunciation gradually grew clear and accurate, and this gradual learning process from "gibberish" to "clear speech" was recorded as audio and played at academic conferences, letting many researchers present feel for the first time, intuitively, that a neural network really has the ability to learn complex rules from data on its own, this demonstration thereafter one of the more widely told classic anecdotes in the history of neural-network research.

Relationships

Ally

🧠Geoffrey Hinton

Echoes today

Below are how modern works borrow or reinterpret this name or story — not the original material. The two differ, so keep them apart.

The "learning to read aloud" demonstration of NETtalkThe recording of NETtalk’s training process, gradually learning clear pronunciation from vague noise, thereafter became one of the more widely circulated classic demonstration cases in the history of neural-network research.

Appears in

The Perceptron, the Winter, and the Revival of ConnectionismThe Rediscovery of the Backpropagation Algorithm

← Back to Figures