Data Processing Using Artificial Neural Networks

The artificial neural network (ANN) is a machine learning (ML) methodology that evolved and developed from the scheme of imitating the human brain. Artificial intelligence (AI) pyramid illustrates the evolution of ML approach to ANN and leading to deep learning (DL). Nowadays, researchers are very much attracted to DL processes due to its ability to overcome the selectivity-invariance problem. In this chapter, ANN has been explained by discussing the network topology and development parameters (number of nodes, number of hidden layers, learning rules and activated function). The basic concept of node and neutron has been explained, with the help of diagrams, leading to the ANN model and its operation. All the topics have been discussed in such a scheme to give the reader the basic concept and clarity in a sequential way from ANN perceptron model to deep learning models and underlying types.

On a Functional Equation Appearing on the Margins of a Mean Invariance Problem

Given a continuous strictly monotonic real-valued function α, defined on an interval I, and a function ω : I → (0, +∞) we denote by Bαω the Bajraktarević mean generated by α and weighted by ω:B_\omega ^\alpha \left({x,y} \right) = {\alpha ^{- 1}}\left({{{\omega \left(x \right)} \over {\omega \left(x \right) + \omega \left(y \right)}}\alpha \left(x \right) + {{\omega \left(y \right)} \over {\omega \left(x \right) + \omega \left(y \right)}}\alpha \left(y \right)} \right),\,\,\,x,y \in I.We find a necessary integral formula for all possible three times differentiable solutions (φ, ψ) of the functional equationr\left(x \right)B_s^\varphi \left({x,y} \right) + r\left(y \right)B_t^\psi \left({x,y} \right) = r\left(x \right)x + r\left(y \right)y,where r, s, t : I → (0, +∞) are three times differentiable functions and the first derivatives of φ, ψ and r do not vanish. However, we show that not every pair (φ, ψ) given by the found formula actually satisfies the above equation.

Mixture multigroup factor analysis for unraveling factor loading non-invariance across many groups

Psychological research often builds on between-group comparisons of (measurements of) latent constructs; for instance, to evaluate cross-cultural differences in neuroticism or mindfulness. A critical assumption in such comparative research is that the same construct(s) are measured in exactly the same way across all groups (i.e., measurement invariance). Otherwise, one would be comparing apples and oranges. Nowadays, measurement invariance is often tested across a large number of groups. When the assumption is untenable, one may compare group-specific measurement models to pinpoint sources of non-invariance, but the number of pairwise comparisons exponentially increases with the number of groups. This makes it hard to unravel invariances from non-invariances and for which groups they apply, and it elevates the chances of falsely detecting non-invariance. An intuitive solution is clustering the groups into a few clusters based on the measurement model parameters. Not only does this confine the number of comparisons needed to identify non-invariances, but the clustering of the groups is an interesting result in itself and provides clues on how to move forward with the between-group comparisons or further measurement invariance testing. To this aim, we present mixture multigroup factor analysis which accommodates a unique blend of cluster- and group-specific parameters to disentangle different levels of non-invariance and set aside parameter differences that are irrelevant to the measurement invariance problem.

EARSHOT: A minimal neural network model of incremental human speech recognition

Despite the “lack of invariance problem” (multiple acoustic patterns map to the same phoneme, and one acoustic pattern can map to different phonemes), humans experience phonetic constancy: we typically perceive what the speaker intended despite this variability. Computational models of human speech recognition have deferred the problem, working with idealized inputs rather than speech. Deep learning models have allowed automatic speech recognition to become robust, but are too complex to relate directly to theories of human speech recognition. We report results from a simple network using long short-term memory nodes, which allow it to learn to recognize real speech with high accuracy, with minimal complexity. Representations emerge in the model that resemble those observed in human cortical responses to speech.

Direct Perception of Speech

The theory of speech perception as direct derives from a general direct-realist account of perception. A realist stance on perception is that perceiving enables occupants of an ecological niche to know its component layouts, objects, animals, and events. “Direct” perception means that perceivers are in unmediated contact with their niche (mediated neither by internally generated representations of the environment nor by inferences made on the basis of fragmentary input to the perceptual systems). Direct perception is possible because energy arrays that have been causally structured by niche components and that are available to perceivers specify (i.e., stand in 1:1 relation to) components of the niche. Typically, perception is multi-modal; that is, perception of the environment depends on specifying information present in, or even spanning, multiple energy arrays. Applied to speech perception, the theory begins with the observation that speech perception involves the same perceptual systems that, in a direct-realist theory, enable direct perception of the environment. Most notably, the auditory system supports speech perception, but also the visual system, and sometimes other perceptual systems. Perception of language forms (consonants, vowels, word forms) can be direct if the forms lawfully cause specifying patterning in the energy arrays available to perceivers. In Articulatory Phonology, the primitive language forms (constituting consonants and vowels) are linguistically significant gestures of the vocal tract, which cause patterning in air and on the face. Descriptions are provided of informational patterning in acoustic and other energy arrays. Evidence is next reviewed that speech perceivers make use of acoustic and cross modal information about the phonetic gestures constituting consonants and vowels to perceive the gestures. Significant problems arise for the viability of a theory of direct perception of speech. One is the “inverse problem,” the difficulty of recovering vocal tract shapes or actions from acoustic input. Two other problems arise because speakers coarticulate when they speak. That is, they temporally overlap production of serially nearby consonants and vowels so that there are no discrete segments in the acoustic signal corresponding to the discrete consonants and vowels that talkers intend to convey (the “segmentation problem”), and there is massive context-sensitivity in acoustic (and optical and other modalities) patterning (the “invariance problem”). The present article suggests solutions to these problems. The article also reviews signatures of a direct mode of speech perception, including that perceivers use cross-modal speech information when it is available and exhibit various indications of perception-production linkages, such as rapid imitation and a disposition to converge in dialect with interlocutors. An underdeveloped domain within the theory concerns the very important role of longer- and shorter-term learning in speech perception. Infants develop language-specific modes of attention to acoustic speech signals (and optical information for speech), and adult listeners attune to novel dialects or foreign accents. Moreover, listeners make use of lexical knowledge and statistical properties of the language in speech perception. Some progress has been made in incorporating infant learning into a theory of direct perception of speech, but much less progress has been made in the other areas.

A Chimpanzee’s (Pan troglodytes) Perception of Variations in Speech: Identification of Familiar Words When Whispered and When Spoken by a Variety of Talkers

When humans perceive speech they process the acoustic properties of the sounds. The acoustics of a specific word can be different depending on who produces it and how they produce it. For example, a whispered word has different acoustic properties than a word spoken in a more natural manner; basically, the acoustics are “noisier.” A word will also sound differently depending on who speaks it, due to the different physical and physiological characteristics of the talker. In this instance, humans routinely normalize speech to retrieve the lexical meaning by solving what is termed the “lack of invariance” problem. We investigated these speech perception phenomena in a language-trained chimpanzee (Pan troglodytes) named Panzee to ascertain if more generalized auditory capabilities, as opposed to specialized human cognitive processes, were adequate to accomplish these perceptual tasks. In Experiment 1 we compared the chimpanzee’s performance when identifying words she was familiar with in natural versus whispered form. In Experiment 2 we investigated Panzee’s ability to solve the “lack of invariance” problem when familiar words were spoken by a variety of talkers (familiar and unfamiliar male and female adults, and children). The results of Experiment 1 demonstrated that there was no difference in her recognition for the two word types. The results of Experiment 2 revealed no significant difference in Panzee’s performance across all talker types. Her overall performance suggests that more generalized capabilities are sufficient for solving for uncertainty when processing the acoustics of speech, and instead favor a strong role of early experience.