Modular design patterns for hybrid learning and reasoning systems

The unification of statistical (data-driven) and symbolic (knowledge-driven) methods is widely recognized as one of the key challenges of modern AI. Recent years have seen a large number of publications on such hybrid neuro-symbolic AI systems. That rapidly growing literature is highly diverse, mostly empirical, and is lacking a unifying view of the large variety of these hybrid systems. In this paper, we analyze a large body of recent literature and we propose a set of modular design patterns for such hybrid, neuro-symbolic systems. We are able to describe the architecture of a very large number of hybrid systems by composing only a small set of elementary patterns as building blocks. The main contributions of this paper are: 1) a taxonomically organised vocabulary to describe both processes and data structures used in hybrid systems; 2) a set of 15+ design patterns for hybrid AI systems organized in a set of elementary patterns and a set of compositional patterns; 3) an application of these design patterns in two realistic use-cases for hybrid AI systems. Our patterns reveal similarities between systems that were not recognized until now. Finally, our design patterns extend and refine Kautz’s earlier attempt at categorizing neuro-symbolic architectures.

     

Differences in Effects of Pyrrolizidine Alkaloids on Five Generalist Insect Herbivore Species

The evolution of the diversity in plant secondary compounds is often thought to be driven by insect herbivores, although there is little empirical evidence for this assumption. To investigate whether generalist insect herbivores could play a role in the evolution of the diversity of related compounds, we examined if (1) related compounds differ in their effects on generalists, (2) there is a synergistic effect among compounds, and (3) effects of related compounds differed among insect species. The effects of pyrrolizidine alkaloids (PAs) were tested on five generalist insect herbivore species of several genera using artificial diets or neutral substrates to which PAs were added. We found evidence that structurally related PAs differed in their effects to the thrips Frankliniella occidentalis, the aphid Myzus persicae, and the locust Locusta migratoria. The individual PAs had no effect on Spodoptera exigua and Mamestra brassicae caterpillars. For S. exigua, we found indications for synergistic deterrent effects of PAs in PA mixtures. The relative effects of PAs differed between insect species. The PA senkirkine had the strongest effect on the thrips, but had no effect at all on the aphids. Our results show that generalist herbivores could potentially play a role in the evolution and maintenance of the diversity of PAs.     

Measurement of prolonged fatigue in the working population: Determination of a cutoff point for the Checklist Individual Strength.

In the Netherlands, a large-scale prospective cohort study was started on prolonged fatigue in the working population. The 1st issue that had to be addressed was the determination of a cutoff point for fatigue for use in the working population. Fatigue is measured with the Checklist Individual Strength (CIS), a 20-item self-report questionnaire. This article demonstrates the process of decision making in the determination of the cutoff point. Total CIS scores were calculated, sensitivity and specificity were compared for potential cutoff points, and a receiver operating characteristics analysis was conducted. A CIS total cutoff point for fatigue of >76 was determined, with a specificity of 90% and a sensitivity of 73%. Limitations regarding the use of cutoff points are discussed. It is concluded that the defined cutoff point seems to be appropriate for use in the working population. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Covalently linked Au nanoparticles to a viral vector: potential for combined photothermal and gene cancer therapy

Hyperthermia can be produced by near-infrared laser irradiation of gold nanoparticles present in tumors and thus induce tumor cell killing via a bystander effect. To be clinically relevant, however, several problems still need to be resolved. In particular, selective delivery and physical targeting of gold nanoparticles to tumor cells are necessary to improve therapeutic selectivity. Considerable progress has been made with respect to retargeting adenoviral vectors for cancer gene therapy. We therefore hypothesized that covalent coupling of gold nanoparticles to retargeted adenoviral vectors would allow selective delivery of the nanoparticles to tumor cells, thus feasibilizing hyperthermia and gene therapy as a combinatorial therapeutic approach. For this, sulfo-N-hydroxysuccinimide labeled gold nanoparticles were reacted to adenoviral vectors encoding a luciferase reporter gene driven by the cytomegalovirus promoter (AdCMVLuc). We herein demonstrate that covalent coupling could be achieved, while retaining virus infectivity and ability to retarget tumor-associated antigens. These results indicate the possibility of using adenoviral vectors as carriers for gold nanoparticles.     

Kinetics measurements and in situ Raman spectroscopy of formation of hydrogen–tetrabutylammonium bromide semi-hydrates

The kinetics of formation of H₂–TBAB semi-clathrate hydrates was studied in this work in order to elucidate their potential for H₂ storage. The influence of pressure (5–16 MPa), TBAB concentration (2.6 mol% and 3.7 mol%) and formation method (T-cycle method and T-constant method) on the hydrate nucleation, hydrate growth and H₂ storage capacity was determined. The results showed that kinetics is favored at higher pressures and solute concentrations. Additionally, the hydrate phase formation and dissociation was study for a solution of 2.6 mol% of TBAB in situ by using the Raman spectroscopy technique. The inclusion of H₂ in the semi-hydrate phase was confirmed. The results showed the importance of H₂ mass transfer on the storage capacity of the H₂–TBAB semi-hydrates.     

Order or disorder? Impaired Hebb learning in dyslexia.

The present study offers an integrative account proposing that dyslexia and its various associated cognitive impairments reflect an underlying deficit in the long-term learning of serial-order information, here operationalized as Hebb repetition learning. In nondyslexic individuals, improved immediate serial recall is typically observed when one particular sequence of items is repeated across an experimental session, a phenomenon known as the Hebb repetition effect. Starting from the critical observation that individuals with dyslexia seem to be selectively impaired in cognitive tasks that involve processing of serial order, the present study is the first to test and confirm the hypothesis that the Hebb repetition effect is affected in dyslexia, even for nonverbal modalities. We present a theoretical framework in which the Hebb repetition effect is assumed to be a laboratory analogue of naturalistic word learning, on the basis of which we argue that dyslexia is characterized by an impairment of serial-order learning that affects language learning and processing. (PsycINFO Database Record (c) 2011 APA, all rights reserved)     

Value sensitive design of a virtual assistant for workload harmonization in teams

Cognition, Technology & Work - Uneven workload distributions in teams can lead to suboptimal team performance. This paper therefore describes the design of a virtual assistant that supports...     

Office workers' computer use patterns are associated with workplace stressors

This field study examined associations between workplace stressors and office workers' computer use patterns. We collected keyboard and mouse activities of 93 office workers (68F, 25M) for approximately two work weeks. Linear regression analyses examined the associations between self-reported effort, reward, overcommitment, and perceived stress and software-recorded computer use duration, number of short and long computer breaks, and pace of input device usage. Daily duration of computer use was, on average, 30 min longer for workers with high compared to low levels of overcommitment and perceived stress. The number of short computer breaks (30 s–5 min long) was approximately 20% lower for those with high compared to low effort and for those with low compared to high reward. These outcomes support the hypothesis that office workers' computer use patterns vary across individuals with different levels of workplace stressors.