In his doctoral work of 1958, he wrote on the interaction of
excitons in crystals, coining the term
polariton for a
quasiparticle that appears in
solid-state physics. He wrote: "The polarization field 'particles' analogous to
photons will be called 'polaritons'." From 1959 to 1963, Hopfield and David G. Thomas investigated the exciton structure of
cadmium sulfide from its reflection spectra. Their experiments and theoretical models allowed to understand the optical spectroscopy of
II-VI semiconductor compounds. Condensed matter physicist
Philip W. Anderson reported that John Hopfield was his "hidden collaborator" for his 1961–1970 works on the
Anderson impurity model which explained the
Kondo effect. Hopfield was not included as a co-author in the papers but Anderson admitted the importance of Hopfield's contribution in various of his writings. William C. Topp and Hopfield introduced the concept of norm-conserving
pseudopotentials in 1973. In 1974 he introduced a mechanism for error correction in
biochemical reactions known as
kinetic proofreading to explain the accuracy of
DNA replication. Hopfield published his first paper in neuroscience in 1982, titled "Neural networks and physical systems with emergent collective computational abilities" where he introduced what is now known as
Hopfield network, a type of artificial network that can serve as a
content-addressable memory, made of binary neurons that can be 'on' or 'off'. The 1982 and 1984 papers represent his two most cited works. Together with
David W. Tank, Hopfield developed a method in 1985–1986 for solving discrete optimization problems based on the continuous-time dynamics using a Hopfield network with continuous activation function. The optimization problem was encoded in the interaction parameters (weights) of the network. The effective temperature of the analog system was gradually decreased, as in global optimization with
simulated annealing. Hopfield is one of the pioneers of the
critical brain hypothesis, he was the first to link neural networks with
self-organized criticality in reference to the
Olami–Feder–Christensen model for earthquakes in 1994. In 1995, Hopfield and Andreas V. Herz showed that avalanches in neural activity follow power law distribution associated to earthquakes. The original Hopfield networks had a limited memory, this problem was addressed by Hopfield and Dimitry Krotov in 2016. Large memory storage Hopfield networks are now known as
modern Hopfield networks. == Views on artificial intelligence ==