>            The Russian Ministry of Science and Technologies and
>            Moscow State University of Engineering and Physics
>            organize the Conference  "Neuroinformatics-99"
>            (Moscow, Russia, 20-22, January, 1999).
> 
>             The  round-table discussion with title
>            "Neurocomputing - 10 further on"  will be held
>             during the Conference.
>           (This discussion is supposed to be connected with the
>            discussion which have  been organized during the Pushchino
>            conference "Architecture of Neurocomputers" in 1988 and the
>            talks of participants have been published in  brochure
>           "Discussion on Neurocomputers", V.I.Kryukov (ed.), Pushchino,
>            1988 and also in "Neural Networks-Theory and
>            Architecture", A.Holden and V.I.Kryukov  (eds), Manchester
>            University Press, UK, 1990)
>            The questions to be discussed are the follows.
> 
>           1. What are the most important achievements of neural network
>              theory in last 10 years?
>           2. What new knowledge has been obtained using
>              computational modelling in the understanding of brain
>              functioning in last decade?
>           3. What are the most prominent developments in the hardware
>              implementation  of neurocomputing in last 10 years?
>           4. What are the most impressive practical applications
>              of neural technologies?
>           5. What are the most promising directions for the future
>              development of neurocomputing?
>           6. What are the frontiers of neurocomputing and what
>              other technologies will successfully cooperate with
>              neurocomputing?
> 

My answer to question 5 also serves as my answer to questions 1, 3, and
6.

There is tremendous potential in exploring the connections between
neurocomputing and rapidly-emerging field of quantum computing.  There
have been some exciting breakthoughs in the theory of quantum
computation in the past few years, such as the possibility of computing
in polynomial time computations that are NP when performed with
classical processors.  But almost all of this work has been from the
perspective of implementing traditional, classical forms of computation
(a notable exception is the work of Ajit Narayanan and his colleagues at
the University of Exeter).  Quantum implementations of neural
architetures might yield even more astounding results, including, but
not necessarily limited to, making learnable what is unlearnable with
classical neurocomputers.  Some might emphasise the possible
contributions to understanding the brain and/or consciousness that
quantum neurocomputers might provide, but with respect to these
possibilities I am more sceptical.  However, even if it turns out that
quantum implementations of neural networks are impossible or
impractical, the (classical) network architectures that they inspire
have already been shown to have some generalisation and learning time
advantages.