Improving user experience during the delivery of immersive content is crucial for its success for both the content creators and audience. Creators can express themselves better with multisensory stimulation, while the audience can experience a higher level of involvement. The rapid development of mulsemedia devices provides better access for stimuli such as olfaction and haptics. Nevertheless, due to the required manual annotation process of adding mulsemedia effects, the amount of content available with sensorial effects is still limited. This work introduces an innovative mulsemedia-enhancement solution capable of automatically generating olfactory and haptic content based on 360 video content, with the use of neural networks. Two parallel neural networks are responsible for automatically adding scents to 360 videos: a scene detection network (responsible for static, global content) and an action detection network (responsible for dynamic, local content). A 360 video dataset with scent labels is also created and used for evaluating the robustness of the proposed solution. The solution achieves a 69.19% olfactory accuracy and 72.26% haptics accuracy during evaluation using two different datasets.

Abstract Recent advancements in multielectrode methods and spike-sorting algorithms enable the in vivo recording of the activities of many neurons at a high temporal resolution. These datasets offer new opportunities in the investigation of the biological neural code, including the direct testing of specific coding hypotheses, but they also reveal the limitations of present decoder algorithms. Classical methods rely on a manual feature extraction step, resulting in a feature vector, like the firing rates of an ensemble of neurons. In this paper, we present a recurrent neural-network-based decoder and evaluate its performance on experimental and artificial datasets. The experimental datasets were obtained by recording the auditory cortical responses of rats exposed to sound stimuli, while the artificial datasets represent preset encoding schemes. The task of the decoder was to classify the action potential timeseries according to the corresponding sound stimuli. It is illustrated that, depending on the coding scheme, the performance of the recurrent-network-based decoder can exceed the performance of the classical methods. We also show how randomized copies of the training datasets can be used to reveal the role of candidate spike-train features. We conclude that artificial neural network decoders can be a useful alternative to classical population vector-based techniques in studies of the biological neural code.

Abstract Ordering is a well-established concept in mathematics and also plays an important role in many areas of computer science, wherequasi-orderings, most notablywell-founded quasi-orderingsandwell-quasi-orderings,are of particular interest. This paper deals with quasi-orderings on first-order terms and introduces a new notion of unification based on a special quasi-order, known ashomeomorphic tree embedding. Historically, the development of unification theory began with the central notion ofa most general unifierbased on thesubsumption order. A unifier $\sigma$ is most general, if it subsumes any other unifier $\tau$ , that is, if there is a substitution $\lambda$ with $\tau=_{E}\sigma\lambda$ , where E is an equational theory and $=_{E}$ denotes equality under E. Since there is in general more than one most general unifier for unification problems under equational theories E, calledE-Unification, we have the notion of a complete and minimal set of unifiers under E for a unification problem $\varGamma$ , denoted as $\mu\mathcal{U}\Sigma_{E}(\Gamma)$ . This set is still the basic notion in unification theory today. But, unfortunately, the subsumption quasi-order is not a well-founded quasi-order, which is the reason why for certain equational theories there are solvable E-unification problems, but the set $\mu\mathcal{U}\Sigma_{E}(\Gamma)$ does not exist. They are called type nullary in the unification hierarchy. In order to overcome this problem and also to substantially reduce the number of most general unifiers, we extended the well-knownencompassment order on termsto anencompassment order on substitutions(modulo E). Unification under the encompassment order is calledessential unificationand if $\mu\mathcal{U}\Sigma_{E}(\Gamma)$ exists, then the complete set of essential unifiers $e\mathcal{U}\Sigma_{E}(\Gamma)$ is a subset of $\mu\mathcal{U}\Sigma_{E}(\Gamma)$ . An interesting effect is that many E-unification problems with an infinite set of most general unifiers (under the subsumption order) reduce to a problem with only finitely manyessentialunifiers. Moreover, there are cases of an equational theory E, for which the complete set of most general unifiers does not exist, theminimal and complete set of essential unifiershowever does exist. Unfortunately again, the encompassment order is not a well-founded quasi-ordering either, that is, there are still theories with a solvable unification problem, for which a minimal and complete set of essential unifiers does not exist. This paper deals with a third approach, namely the extension of the well-knownhomeomorphic embedding of termsto ahomeomorphic embedding of substitutions (modulo E). We examine the set of most general, minimal, and complete E-unifiers under the quasi-order of homeomorphic embedment modulo an equational theory E, called $\varphi U\Sigma_{E}(\Gamma)$ , and propose an appropriate definitional framework based on the standard notions of unification theory extended by notions for thetree embedding theoremor Kruskal’s theorem as it is called. The main results are that forregulartheories the minimal and complete set $\varphi\mathcal{U}\Sigma_{E}(\Gamma)$ always exists. If we restrict the E-embedding order topure E-embedding, awell-known technique in logic programming and term rewriting where the difference between variables is ignored, the set $\varphi_{\pi}\mathcal{U}\Sigma_{E}(\Gamma)$ always exists and it is even finite for any theory E.

There is increased awareness of the possibility of developmental memories resulting from evolutionary learning. Genetic regulatory and neural networks can be modelled by analogous formalism raising the important question of productive analogies in principles, processes and performance. We investigate the formation and persistence of various developmental memories of past phenotypes asking how the number of remembered past phenotypes scales with network size, to what extent memories stored form by Hebbian-like rules, and how robust these developmental "devo-engrams" are against networks perturbations (graceful degradation). The analogy between neural and genetic regulatory networks is not superficial in that it allows knowledge transfer between fields that used to be developed separately from each other. Known examples of spectacular phenotypic radiations could partly be accounted for in such terms.

The Octopus arm is a highly versatile and complex limb. How the Octo- pus controls such a hyper-redundant arm (not to mention eight of them!) is as yet unknown. Robotic arms based on the same mechanical prin- ciples may render present day robotic arms obsolete. In this paper, we tackle this control problem using an online reinforcement learning al- gorithm, based on a Bayesian approach to policy evaluation known as Gaussian process temporal difference (GPTD) learning. Our substitute for the real arm is a computer simulation of a 2-dimensional model of an Octopus arm. Even with the simplifications inherent to this model, the state space we face is a high-dimensional one. We apply a GPTD- based algorithm to this domain, and demonstrate its operation on several learning tasks of varying degrees of difficulty.

Abstract We show that the DA-uniflcation problem is undecidable. That is, given two binary function symbols ⊕ and ⊗, variables and constants, it is undecidable if two terms built from these symbols can be unified provided the following DA-axioms hold: Two terms are DA-unifiable (i.e. an equation is solvable in DA) if there exist terms to be substituted for their variables such that the resulting terms are equal in the equational theory DA. This is the smallest currently known axiomatic subset of Hilbert's tenth problem for which an undecidability result has been obtained.