Pre-Grant Publication Number: 20090132452
Please help the USPTO examine the application by evaluating the relevance of the publicly submitted prior art to the patent application.
Peer To Patent forwards the Top 10 most relevant prior art submissions and their annotations to the USPTO.
Review this prior art and click on the thumbs up (or down) to indicate whether this submission should be forwarded to the USPTO.
If you login then you can add an annotation by typing in the box at the bottom of the screen to comment on the relevance of the prior art to the claims of the patent application.
Review this prior art and click on the thumbs up (or down) to indicate whether this submission should be forwarded to the USPTO.
If you login then you can add an annotation by typing in the box at the bottom of the screen to comment on the relevance of the prior art to the claims of the patent application.
Prior Art Detail
Summary / Description
| Summary / Description | Abstract: We present a generalization of conventional artifi cial neural networks that allows for a functional equivalence to multi-expert systems. The new model provides an architectural freedom going beyond existing multi-expert models and an integrative formalism to compare and combine various techniques of learning. (We consider gradient, EM, reinforcement, and unsupervised learning.) Its uniform representation aims at a simple genetic encoding and evolutionary structure optimization of multi-expert systems. This paper contains a detailed description of the model and learning rules, empirically validates its functionality, and discusses future perspectives. |
Basic Information
| Type of Prior Art | Print Publication |
| Publication Title * | Proceedings of the International Joint Conference on Neural Networks (IJCNN 2002) |
| Author | Marc Toussaint |
| ISBN | |
| Page Range | |
| Medium | Journal article |
| Publication Date * | January 29, 2002 |
| URL | |
Notes / To Do
| Notes | Submitted on behalf of Ralph Linsker, IBM, by Diane Willis. |
Excerpt
Excerpt To realize a seperation of the stimulus space one
could rely on the conventional way of implementing
multi-experts, i.e., allow neural networks for the im-
plementation of expert modules and use external, of-
ten more abstract types of gating networks to orga-
nize the interaction between these modules. Much
research is done in this direction (Bengio & Frasconi
1994; Cacciatore & Nowlan 1994; Jordan & Jacobs
1994; Rahman & Fairhurst 1999; Ronco, Gollee, &
Gawthrop 1997). The alternative we want to propose
here is to introduce a neural model that is capable to
represent systems that are functionally equivalent to
multi-expert systems within a single integrative net-
work. This network does not explicitly distinguish
between expert and gating modules and generalizes
conventional neural networks by introducing a coun-
terpart for gating interactions. What is our moti-
vation for such a new representation of multi-expert
systems?
* First, our representation allows much more and
qualitatively new architectural freedom. E.g.,
gating neurons may interact with expert neu-
rons; gating neurons can be a part of experts.
There is no restriction with respect to serial,
parallel, or hierarchical architectures|in a much
more general sense as proposed in (Jordan &
Jacobs 1994).
* Second, our representation allows in an intu-
itive way to combine techniques from various
learning theories. This includes gradient de-
scent, unsupervised learning methods like Hebb
learning or the Oja rule, and an EM-algorithm
that can be transferred from classical gating-
learning theories (Jordan & Jacobs 1994). Fur-
ther, the interpretation of a speci*c gating as
an action exploits the realm of reinforcement
learning, in particular Q-learning and (though
not discussed here) its TD and TD(*) variants
(Sutton & Barto 1998).
* Third, our representation makes a simple ge-
netic encoding of such architectures possible.
There already exist various techniques for evo-
lutionary structure optimization of networks (see
(Yao 1999) for a review). Applied on our repre-
sentation, they become techniques for the evo-
lution of multi-expert architectures.
After the rather straight-forward generalization of
neural interactions necessary to realize gatings (sec-
tion II), we will discuss in detail di*erent learning
methods in section III. The empirical study in sec-
tion IV compares the di*erent interactions and learn-
ing mechanisms on a test problem similar to the one
discussed by Jacobs et al. (1990).
1 |
Relevance
Claims
1
Relevance
This patent deals with the idea of a second computation that determines (or "gates") whether the result of a first computation should be used, or should instead be replaced by a predetermined value, in a neural network, which exists in prior art. For example, the architecture of multiple neural nets whose outputs feed, via a gating process, into a higher stage of the overall network. Only one of the values is passed through the gate at a given time. The one passed may depend on the 'confidence' that each network has in its output result. This patent teaches this principle. See the Excerpt.
This patent deals with the idea of a second computation that determines (or "gates") whether the result of a first computation should be used, or should instead be replaced by a predetermined value, in a neural network, which exists in prior art. For example, the architecture of multiple neural nets whose outputs feed, via a gating process, into a higher stage of the overall network. Only one of the values is passed through the gate at a given time. The one passed may depend on the 'confidence' that each network has in its output result. This patent teaches this principle. See the Excerpt.
Claim Chart
Some
9
Relevance
See the Excerpt and Claim 1.
See the Excerpt and Claim 1.
Claim Chart
Some
18
Relevance
See the Excerpt and Claim 1.
See the Excerpt and Claim 1.
Claim Chart
All
