Fuzzy Logic and Neural Nets: Still Viable?
Posted 13 Jun 2004 at 03:28 UTC by steve
A new EDN
article discusses whether or not Fuzzy Logic and Neural Networks are
still useful engineering tools after all these years, even though their
15 minutes of fame seems to have passed. The article talks about several
companies that still use or sell products based on each technology.
Fuzzy logic and neural networks are methods whose time has yet to come.
Fuzzy type logic is already used in things like speech recognition
software, but neural networks havn't seen much commercial success.
The main problems with neural networks is that most of the algorithms
developed so far require extensive training periods based upon
hand-picked examples. Where neural networks do make a difference is in
things like face recognition, recognising different types of vehicle
travelling along a road, or really any situation involving visual
identification. Although vision systems have been used extensively in
industry they havn't yet broken out into the mainstream consumer market.
As digital cameras and processing power becomes ever cheaper I think
that situation will change over the next 5 years, and we will start to
see things like childrens toys with face and voice recognition technology.
Yeah but, posted 13 Jun 2004 at 11:17 UTC by c6jones720 »
I agree with Motters that ANNs and Fuzzy are viable but still need a
little more proving. The only problem is that it seems that there are
very few people who actually understand the concepts and are able to
communicate them to other people. I am very interested in these things
but was not convinced of my lecturers understanding of these ares when
they tried to explain the workings to our group. I do believe these
have futures, but may need a little more testing first.
I agree on neural nets. I think they have a lot of potential that has
yet to be recognized and developed. I still see a lot of activity
involving neural nets on both the research side and in practical
applications. They're still in their infancy.
Fuzzy logic, on the other hand, has turned out to be something of a
bust. It's not that it didn't work. But it's just a fancy name for
one particular method of doing something that engineers had been doing
for years anyway. Bob Pease wrote a series of pretty damning articles in
which he takes fuzzy logic to task and challenges its claimed
virtues, showing that traditional engineering methods beat it in every case
where he could get enough verifiable data to build a conventional
controller. (which wasn't many as most of the well-known uses of fuzzy
logic hyped by proponents turned out to be bogus or highly exagerated).
I was going to link them to the above article but
could only find three of the five online. Here are the ones that I could
What's all this Fuzzy Logic Stuff, Anyhow? Part I - Anyone have a link??
In this one he first stated his problems with fuzzy logic hype and
proponents to provide any evidence - in a form he could reproduce in the
lab, such as schematics or code - that fuzzy logic could do anything
better than conventional techniques like PID.
this Fuzzy Logic Stuff, Anyhow? Part II
What's all this Fuzzy Logic Stuff, Anyhow? Part III (link?)
all this Fuzzy Logic Stuff, Anyhow? Part IV
all this Fuzzy Logic Stuff, Anyhow? Part V
thoughts on Fuzzy logic
And one other related article:
Crisp Thoughts on Fuzzy Logic by Daniel Abramovitch of HP Labs.
Binary is King, posted 13 Jun 2004 at 17:18 UTC by ROB.T. »
With binary logic you can emulate fuzzy logic and neural networks,
which makes them a subset of the full capabilities of binary. From
what I can see the only advantages of having true fuzzy/neural
circuits is to save space/time in your design, and I'm not sure those
advantages are worth it for any design.
More then less, posted 15 Jun 2004 at 03:32 UTC by roschler »
Fuzzy Logic, within the realm of control electronics and process
control, is actually an established and flourishing technology,
especially in Japan. Outside of the that dual niche it gets a lot more
Neural nets are also a niche dominant technology. Credit card fraud
systems have been in use for many years using neural net technology, and
many popular OCR systems use them.
It's more of a matter of finding a conducive problem space with Fuzzy
Logic, one that fits the solution. With neural nets, the other posters
comments about the difficulty in using them come into play. You need a
high skill level in many different fields, especially statistics and
data mining, to really make them fly.
A friend and I were talking the other day about how to design a control
algorithm that could fly a small autonomous plane through a forest.
Basiaclly, we figured, a dragonfly can do a great job. Yet a dragonfly
has hardly any brain at all.
Clearly the dragonfly's ability to navigate a forest is not based on
complex algorithms, but some pretty tight wiring between eyes and wings.
"Something close to the left, shift right!" etc.
Neural nets and fuzzy sets seem a much better way to imagine how you'd
dragonflies, posted 15 Jun 2004 at 20:45 UTC by steve »
Yep, I'm always fascinated by dragonflies and insect-level intelligence
in general. I've seen several optical flow implementations that use
neural networks. Optical flow is essentially what dragflies (and other
insects with compound eyes) are using. The output of a well-designed
optical flow vision system provides vector based information very
similar to what you'd get from the gyro/accelerometer combination in an
IMU. It can tell you yaw, roll, and pitch accelerations and velocities
as well as heading information; all without traditional, complex vision
processing algorithms. It's subject to certain types of optical
illusions in much the same way
human vision is but it seems to work pretty well for dragonflies...
I disagree that "there are very few people who actually understand the
concepts". Maybe they're not being communicated to a wide audience very
well but the concepts behind Neural Nets are very well understood by
And I strongly agree with some of the negative comments about Fuzzy
Logic. While I'm not a fan of Neural Nets I think they do have a place
in many people's kits of parts. On the other hand Fuzzy Logic is a
completely trivial subject. It is somethng that has been reinvented many
times by many people because it is so simple and many of those people
move onto more sophisticated methods when they eventually learn them.