Reading What Machines “Think” - Brain Informatics - page 169

either between the machine learning algorithms or between the quality of the
images. Programming languages produce different organizations of the memory
of the processes.
Deciding the algorithm performed by a process is instead more complex. Yet,
results are encouraging. The performance of the decision tree learning algorithm
is above 80%. The performance is even higher for smoothed images. The differ-
ences between the naive bayes algorithm and the decision tree learning suggests
that some features are more informative than others. These features are even
more important for smoothed images. This increase in performance is unex-
pected. The behavior of the naive bayes algorithm is instead more predictable as
classifiers learnt and applied on plain images performed better than those learnt
and applied on smoothed images.
These are initial answers to the two questions issued in Sec. 2. As discussed in
the next section, we still need to investigate more complex datasets to generilze
these initial findings.

6 Conclusions and Future Work

We introduced a new vision that can bothhelpinansweringquestionsinneu-
roimaging and produce novel applications in the computer field. The results of
the experimental evaluation suggest we can read what computers “ think ”aswe
can positively predict “ cognitive processes ” from activation images. Then, we can
have positive expectations of the foundational perspective and the application
perspective. Yet, the research program we started is still at the beginning and
many questions still have to be addressed. For this study we investigated simple
“ cognitive processes ”. We need to scale-up to different datasets, e.g., datasets
containing activation images of word processors and image processors. Then,
we have to attack the problem of determining which processes are active in a
given memory. Finally, we have to figure out physical devices to directly capture
activation images from the electronic computer.

Acknowledgement

We would like to thank Marco Cesati for his precious advices on the organization
of the memory in the Linux kernel.

References

1. Wiener, N.: Cybernetics: Or the Control and Communication in the Animal and
the Machine. MIT Press, Cambridge (1948)
2. von Neumann, J.: First draft of a report on the edvac. IEEE Ann. Hist. Com-
put. 15(4), 27-75 (1993)
3. Turing, A.: Intelligent Machinery. In: Meltzer, B., Michie, D. (eds.) Machine Intel-
ligence, vol. 5, pp. 3-23. Edinburgh University Press, Edinburgh (1969)
4. Neisser, U.: Cognitive Psychology. Appleton-Century-Crofts, New York (1967)