Abstract
In this paper we present a connectionist architecture called C-block combining several powerful and cognitively relevant features. It can learn sequential dependencies in incoming data and predict probability distributions over possible next inputs, notice repeatedly occurring sequences, automatically detect sequence boundaries (based on surprise in prediction) and represent sequences declaratively as chunks/plans for future execution or replay. It can associate plans with reward, and also with their effects on the system state. It also supports plan inference from an observed sequence of behaviours: it can recognize possible plans, along with their likely intended effects and expected reward, and can revise these inferences as the sequence unfolds. Finally, it implements goal-driven behaviour, by finding and executing a plan that most effectively reduces the difference between the current system state and the agent’s desired state (goal). C-block is based on modified self-organizing maps that allow fast learning, approximate queries and Bayesian inference.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
This happens outside the C-block proper. ‘Reafferent’ representations of executed actions are passed back to the system as perceptual inputs at the next time step.
- 2.
More formally, we can write this \(\text{ A }+c\cdot \text{ J }+c^2\cdot \text{ prev }\), where ‘prev’ is whatever preceded J, and \(c<1\) is a decay coefficient, usually between 0.5–0.9.
- 3.
A similar approach to encoding chunks is used in [8], with the difference that they use a SRN instead of an SOM and their tonic representation follows a more complex update rule and is more noisy.
- 4.
The distribution is encoded by the SOM activity. We can pick the most active plan, or reconstruct its expected value based on the whole distribution (see Appendix for details).
- 5.
Actually, what is produced is a sequence of probability distributions over possible inputs, from each of which a 1-hot input is selected.
- 6.
The remaining 74 components were zero all the time. This was because we used the same parameters and network sizes across several applications, in some of which we needed more neurons to encode the input.
- 7.
We did this for all but the first position in a chunk, because C-block has no means to predict the continuation of a chunk before it has started.
- 8.
Predicting the winner with probability \(p=1\) leads to KLD = 0. Flat predicted distribution \(p=1/26\) leads to KLD = 1.
- 9.
c regulates the width of the Gaussian. We used \(c=30\) for sequencing SOM and varied \(c=\{15,25,2\}\) for plan SOM predicting from Tonic, Reward and Effect respectively.
References
Bar, M.: Predictions: a universal principle in the operation of the human brain. Philos. Trans. R. Soc. Series B 364(1521), 1181–1182 (2009)
Elman, J.: Finding structure in time. Cogn. Sci. 14, 179–211 (1990)
Friston, K.: The free-energy principle: a unified brain theory? Nature Rev. Neurosci. 11, 127–138 (2010)
Graybiel, A.: The basal ganglia and chunking of action repertoire. Neurobiol. Learn. Mem. 70(1–2), 119–136 (1998)
Kohonen, T.: Self-organized formation of topologically correct feature maps. Biol. Cybernet. 43, 59–69 (1982)
Kurby, C., Zacks, J.: Segmentation in the perception and memory of events. Trends Cogn. Sci. 12(2), 72–79 (2007)
Parr, T., Friston, K.: The anatomy of inference: Generative models and brain structure. Front. Comput. Neurosci. 12, 90 (2018)
Reynolds, J., Zacks, J., Braver, T.: A computational model of event segmentation from perceptual prediction. Cogn. Sci. 31, 613–643 (2007)
Sagar, M., Seymour, M., Henderson, A.: Creating connection with autonomous facial animation. Commun. ACM 59(12), 82–91 (2016)
Strickert, M., Hammer, B.: Merge SOM for temporal data. Neurocomputing 64, 39–71 (2005)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
A Bayesian Inference in the SOM
A Bayesian Inference in the SOM
In our version of SOM, the activity \(A_i\) of each unit is computed as
\(d^2\!\left( \mathbf {x},\mathbf {w}_i\right) \) is the squared Euclidean distance between the input \(\mathbf {x}\) and the weight vector \(\mathbf {w}_i\), \(a_i\) is the (unnormalized) activity of the i-th unit, \(m_i\) is the activation mask for the i-th unit. Activities \(A_i\) are normalized to sum to 1.
Comparing Eq. 1 with the standard Bayes’ rule
we can interpret the activity of each unit as the posterior probability \(p(h_i|d)\) of the hypothesis that the current SOM input (data) belongs to the class represented by the unit i. The GaussianFootnote 9 term \(\exp \left( -c \cdot d^2\!\left( \mathbf {w}_i,\mathbf {x}\right) \right) \) corresponds to the likelihood \(p(d|h_i)\). The mask \(m_i\) corresponds to the prior probability of the i-th hypothesis \(p(h_i)\). The denominator \(\sum _{j=1}^N{a_j}\) in the formula for normalized activities \(A_i\) is a total response of the map to the current input and corresponds to \(\sum _{j=1}^p(d|h_j)\cdot p(h_j)=p(d)\), which is just the probability of the data itself. A very low total activity in the map indicates strange (or novel) input data.
By specifying coefficients \(m_i\), we can choose different prior bias on the SOM, for example relative frequency of how often the i-th neuron became the best-matching unit in the past. All \(m_i\) equal to the same value would effectively mean a uniform prior and will have no influence.
Normalized activity of the whole SOM corresponds to the posterior probability distribution over all the hypotheses/neurons given the current input/data. We can reconstruct the most likely input either as the weights of the winner (‘hard’ output) or as an activity-weighted combination of the weights of all the neurons (‘soft’ output): \(\mathbf {y}=\sum _{j=1}^N{A_j\cdot \mathbf {w}_j}\), which corresponds to the expected value of the input given the distribution.
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Takac, M., Knott, A., Sagar, M. (2020). SOM-Based System for Sequence Chunking and Planning. In: Farkaš, I., Masulli, P., Wermter, S. (eds) Artificial Neural Networks and Machine Learning – ICANN 2020. ICANN 2020. Lecture Notes in Computer Science(), vol 12396. Springer, Cham. https://doi.org/10.1007/978-3-030-61609-0_53
Download citation
DOI: https://doi.org/10.1007/978-3-030-61609-0_53
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-61608-3
Online ISBN: 978-3-030-61609-0
eBook Packages: Computer ScienceComputer Science (R0)