Quicknet is the software library used as the core of all the recognition training currently performed at ICSI. This year it has essentially been in a maintenance phase, fixing any bugs that arose (none serious). We have also begun integrating support for the MultiSPERT systems described above.
Another area of work between ourselves and Cambridge has been the effort to
support recursive neural network (RNN) training on the SPERT boards via
Quicknet. Despite various approaches, there remain problems with this
system, apparently arising from the difference between fixed- and floating-