Learning with incomplete information and the mathematical structure behind it

We investigate the problem of learning with incomplete information as exemplified by learning with delayed reinforcement. We study a two phase learning scenario in which a phase of Hebbian associative learning based on momentary internal representations is supplemented by an ‘unlearning’ phase depending on a graded reinforcement signal. The reinforcement signal quantifies the success-rate globally for a number of learning steps in phase one, and ‘unlearning’ is indiscriminate with respect to associations learnt in that phase. Learning according to this model is studied via simulations and analytically within a student–teacher scenario for both single layer networks and, for a committee machine. Success and speed of learning depend on the ratio λ of the learning rates used for the associative Hebbian learning phase and for the unlearning-correction in response to the reinforcement signal, respectively. Asymptotically perfect generalization is possible only, if this ratio exceeds a critical value λ _c , in which case the generalization error exhibits a power law decay with the number of examples seen by the student, with an exponent that depends in a non-universal manner on the parameter λ. We find these features to be robust against a wide spectrum of modifications of microscopic modelling details. Two illustrative applications—one of a robot learning to navigate a field containing obstacles, and the problem of identifying a specific component in a collection of stimuli—are also provided.     

Hebbian errors in learning: An analysis using the Oja model

Background: Recent work on long term potentiation in brain slices shows that Hebb's rule is not completely synapse-specific, probably due to intersynapse diffusion of calcium or other factors. We previously suggested that such errors in Hebbian learning might be analogous to mutations in evolution.Methods and findings: We examine this proposal quantitatively, extending the classical Oja unsupervised model of learning by a single linear neuron to include Hebbian inspecificity. We introduce an error matrix E, which expresses possible crosstalk between updating at different connections. When there is no inspecificity, this gives the classical result of convergence to the first principal component of the input distribution (PC1). We show the modified algorithm converges to the leading eigenvector of the matrix EC, where C is the input covariance matrix. In the most biologically plausible case when there are no intrinsically privileged connections, E has diagonal elements Q and off-diagonal elements (1-Q)/(n-1), where Q, the quality, is expected to decrease with the number of inputs n and with a synaptic parameter b that reflects synapse density, calcium diffusion, etc. We study the dependence of the learning accuracy on b, n and the amount of input activity or correlation (analytically and computationally). We find that accuracy increases (learning becomes gradually less useful) with increases in b, particularly for intermediate (i.e., biologically realistic) correlation strength, although some useful learning always occurs up to the trivial limit Q=1/n.Conclusions and significance: We discuss the relation of our results to Hebbian unsupervised learning in the brain. When the mechanism lacks specificity, the network fails to learn the expected, and typically most useful, result, especially when the input correlation is weak. Hebbian crosstalk would reflect the very high density of synapses along dendrites, and inevitably degrades learning.     

Doing Philosophy Effectively: Student Learning in Classroom Teaching

An important aim of teaching philosophy in Dutch secondary schools is to learn about philosophy (i.e., the great philosophers) by doing philosophy. We examined doing philosophy and focused specifically on the relationship between student learning activities and teacher behavior; in doing so, a qualitative cross-case analysis of eight philosophy lessons was performed. The effectiveness of doing philosophy was operationalized into five learning activities comprising rationalizing, analyzing, testing, producing criticism, and reflecting, and scored by means of qualitative graphical time registration. Using CA we find a quantitative one-dimensional scale for the lessons that contrasts lessons that are more and less effective in terms of learning and teaching. A relationship was found between teaching by teachers and doing philosophy by students. In particular we found students to produce a higher level of doing philosophy with teachers who chose to organize a philosophical discussion with shared guidance by the teacher together with the students.     

Social learning improves survivorship at a life-history transition

During settlement, one of the main threats faced by individuals relates to their ability to detect and avoid predators. Information on predator identities can be gained either through direct experience or from the observation and/or interaction with others, a process known as social learning. In this form of predator recognition, less experienced individuals learn from experienced members within the social group, without having to directly interact with a predator. In this study, we examined the role of social learning in predator recognition in relation to the survival benefits for the damselfish, Pomacentrus wardi, during their settlement transition. Specifically, our experiments aimed to determine if P. wardi are capable of transmitting the recognition of the odour of a predator, Pseudochromis fuscus, to conspecifics. The experiment also examined whether there was a difference in the rate of survival between individuals that directly learnt the predator odour and those which acquired the information through social learning compared to naïve individuals. Results show that naïve P. wardi are able to learn a predator’s identity from experienced individuals via social learning. Furthermore, survival between individuals that directly learnt the predator’s identity and those that learnt through social learning did not significantly differ, with fish from both treatments surviving at least five times better than controls. These results demonstrate that experience may play a vital role in determining the outcome of predator–prey interactions, highlighting that social learning improves the ability of prey to avoid and/or escape predation at a life-history transition.     

The role of medial temporal lobe structures in implicit learning: an event-related FMRI study

The medial temporal lobe (MTL) has been associated with declarative learning of flexible relational rules and the basal ganglia with implicit learning of stimulus-response mappings. It remains an open question of whether MTL or basal ganglia are involved when learning flexible relational contingencies without awareness. We studied learning of an explicit stimulus-response association with fMRI. Embedded in this explicit task was a hidden structure that was learnt implicitly. Implicit learning of the sequential regularities of the "hidden rule" activated the ventral perirhinal cortex, within the MTL, whereas learning the fixed stimulus-response associations activated the basal ganglia, indicating that the function of the MTL and the basal ganglia depends on the learned material and not necessarily on the participants' awareness.     

The kinetic and chemical mechanism of high-fidelity DNA polymerases

This review summarizes our current understanding of the structural, kinetic and thermodynamic basis for the extraordinary accuracy of high-fidelity DNA polymerases. High-fidelity DNA polymerases, such as the enzyme responsible for the replication of bacteriophage T7 DNA, discriminate against similar substrates with an accuracy that approaches one error in a million base pairs while copying DNA at a rate of approximately 300 base pairs per second. When the polymerase does make an error, it stalls, giving time for the slower proofreading exonuclease to remove the mismatch so that the overall error frequency approaches one in a billion. Structural analysis reveals a large change in conformation after nucleotide binding from an open to a closed state. Kinetic analysis has shown that the substrate-induced structural change plays a key role in the discrimination between correct and incorrect base pairs by governing whether a nucleotide will be retained and incorporated or rapidly released.     

Optimization of Annealing Temperature To Reduce Bias Caused by a Primer Mismatch in Multitemplate PCR

To reduce PCR bias derived from a primer mismatch, the effect of the annealing temperature on the product ratio was investigated by denaturing gradient gel electrophoresis analysis of PCR products from a mixture of perfect-match and one-mismatch templates. These templates were generated by PCR from Pediococcus acidilactici for one mismatch and Micrococcus luteus for the perfect match. PCRs showed that the bias was reduced at lower temperatures. An environmental sample was also examined.     

Emergence of symmetric, modular, and reciprocal connections in recurrent networks with Hebbian learning

While learning and development are well characterized in feedforward networks, these features are more difficult to analyze in recurrent networks due to the increased complexity of dual dynamics – the rapid dynamics arising from activation states and the slow dynamics arising from learning or developmental plasticity. We present analytical and numerical results that consider dual dynamics in a recurrent network undergoing Hebbian learning with either constant weight decay or weight normalization. Starting from initially random connections, the recurrent network develops symmetric or near-symmetric connections through Hebbian learning. Reciprocity and modularity arise naturally through correlations in the activation states. Additionally, weight normalization may be better than constant weight decay for the development of multiple attractor states that allow a diverse representation of the inputs. These results suggest a natural mechanism by which synaptic plasticity in recurrent networks such as cortical and brainstem premotor circuits could enhance neural computation and the generation of motor programs. Received: 27 April 1998 / Accepted in revised form: 16 March 1999     

Discontinuous Molecular Dynamics (DMD) Study of Heteropolymer Collapse in an Explicit Solvent

In this study, we employ the discontinuous molecular dynamics simulation method to investigate the collapse properties of a single heteropolymer chain in an explicit solvent. Solvent density ρ, fraction of hydrophobic monomers n _H (defined as the ratio of the number of hydrophobic monomers to the total number of monomers) and a hydrophobicity parameter λ (which controls the energy mismatch between the monomers and solvent particles) were systematically varied to examine their role in polymer collapse. The average static structure factor of the polymer was used to find the so-called θ-point characterizing the state of an ideal chain. Phase diagrams of ρ versus λ for the coil–globule transition were mapped out for different values of n _H. Increasing the fraction of hydrophobic monomers n _H, solvent density ρ, and hydrophobicity parameter λ were all shown to aid in stabilizing the globule phase. In an effort to explore scaling behaviour of the coil–globule phase diagram as a function of n _H, and to investigate whether the phase boundaries for different n _H collapsed on to one universal curve, we rescaled λ by n _H ^δ λ; we determined δ = 1.72, in contrast to mean-field predictions of δ = 2.0.     

Population heterogeneity in clinical cohorts affects the predictive accuracy of brain imaging

Brain imaging research enjoys increasing adoption of supervised machine learning for single-participant disease classification. Yet, the success of these algorithms likely depends on population diversity, including demographic differences and other factors that may be outside of primary scientific interest. Here, we capitalize on propensity scores as a composite confound index to quantify diversity due to major sources of population variation. We delineate the impact of population heterogeneity on the predictive accuracy and pattern stability in 2 separate clinical cohorts: the Autism Brain Imaging Data Exchange (ABIDE, n = 297) and the Healthy Brain Network (HBN, n = 551). Across various analysis scenarios, our results uncover the extent to which cross-validated prediction performances are interlocked with diversity. The instability of extracted brain patterns attributable to diversity is located preferentially in regions part of the default mode network. Collectively, our findings highlight the limitations of prevailing deconfounding practices in mitigating the full consequences of population diversity.

Brain-imaging research enjoys increasing adoption of supervised machine learning for single-subject disease classification. This study explores the contribution of diversity-aware machine learning models to tracking, unpacking and understanding out-of-distribution generalization in large-scale neuroimaging datasets, and shows that population diversity is a key factor contributing to generalization performance.     

Measuring the Quality of University Lectures: Development and Validation of the Instructional Skills Questionnaire (ISQ)

In higher education, student ratings are often used to evaluate and improve the quality of courses and professors’ instructional skills. Unfortunately, student-rating questionnaires rarely generate specific feedback for professors to improve their instructional skills. The impact of student ratings on professors’ instructional skills has proven to be low. This study concerns the psychometric properties of the Instructional Skills Questionnaire (ISQ), a new theory-based student-rating-of-teaching questionnaire with specific questions concerning lecturing skills. The ISQ is administered after a single lecture. This way, it serves as a formative feedback instrument for university professors during courses to assist them to improve and (re-) evaluate their skills if necessary. The ISQ contains seven dimensions of professors’ instructional skills and three student (self perceived) learning outcomes. In this study, Dutch students in 75 courses rated three 90-minute lectures (T1, T2 and T3) of their respective professors using the ISQ. In total, 14,298 ISQ-forms were used to rate 225 lectures. The teacher level reliabilities of the seven dimensions were found to be good at each measurement occasion. In addition, confirmatory multilevel factor analysis confirmed a seven dimensional factor structure at the teacher level at each measurement occasion. Furthermore, specific teacher level factors significantly predicted students’ (self-assessed) learning outcomes. These results partly supported the proposed theoretical framework on the relationship between the ISQ teaching dimensions and the student learning process, and provided evidence for the construct validity of the instrument. In sum, the ISQ is found to be a reliable and valid instrument, which can be used by professors and faculty development centers to assess and improve university teaching.     

DNA multiplex hybridization on microarrays and thermodynamic stability in solution: a direct comparison

Hybridization intensities of 30 distinct short duplex DNAs measured on spotted microarrays, were directly compared with thermodynamic stabilities measured in solution. DNA sequences were designed to promote formation of perfect match, or hybrid duplexes containing tandem mismatches. Thermodynamic parameters ΔH°, ΔS° and ΔG° of melting transitions in solution were evaluated directly using differential scanning calorimetry. Quantitative comparison with results from 63 multiplex microarray hybridization experiments provided a linear relationship for perfect match and most mismatch duplexes. Examination of outliers suggests that both duplex length and relative position of tandem mismatches could be important factors contributing to observed deviations from linearity. A detailed comparison of measured thermodynamic parameters with those calculated using the nearest-neighbor model was performed. Analysis revealed the nearest-neighbor model generally predicts mismatch duplexes to be less stable than experimentally observed. Results also show the relative stability of a tandem mismatch is highly dependent on the identity of the flanking Watson–Crick (w/c) base pairs. Thus, specifying the stability contribution of a tandem mismatch requires consideration of the sequence identity of at least four base pair units (tandem mismatch and flanking w/c base pairs). These observations underscore the need for rigorous evaluation of thermodynamic parameters describing tandem mismatch stability.     

Application of spectrally resolved fluorescence induction to study light-induced nonphotochemical quenching in algae

The light-induced nonphotochemical quenching (NPQ) can safely dissipate excess of absorbed light to heat. Here we describe an application of spectrally resolved fluorescence induction (SRFI) method for studying spectral variability of NPQ. The approach allows detection of spectrally-resolved nonphotochemical quenching (NPQ_λ) representing NPQ dependency on fluorescence emission wavelength in the whole spectral range of fluorescence emission. The experimental approach is briefly described and NPQ_λ is studied for the cryptophyte alga Rhodomonas salina and for green alga Chlorella sp. We confirm presence of NPQ_λ only in membrane-bound antennae (chlorophyll a/c antennae) and not in phycobiliproteins in lumen in cryptophyte and show that NPQ_λ is inhibited in the whole spectral range by NPQ inhibitors in Chlorella sp. We discuss variability in the quenching in the particular spectral ranges and applicability of the NPQ_λ parameter to study quenching locus in vivo.     

Learning outdoors: male lizards show flexible spatial learning under semi-natural conditions

Spatial cognition is predicted to be a fundamental component of fitness in many lizard species, and yet some studies suggest that it is relatively slow and inflexible. However, such claims are based on work conducted using experimental designs or in artificial contexts that may underestimate their cognitive abilities. We used a biologically realistic experimental procedure (using simulated predatory attacks) to study spatial learning and its flexibility in the lizard Eulamprus quoyii in semi-natural outdoor enclosures under similar conditions to those experienced by lizards in the wild. To evaluate the flexibility of spatial learning, we conducted a reversal spatial-learning task in which positive and negative reinforcements of learnt spatial stimuli were switched. Nineteen (32%) male lizards learnt both tasks within 10 days (spatial task mean: 8.16 ± 0.69 (s.e.) and reversal spatial task mean: 10.74 ± 0.98 (s.e.) trials). We demonstrate that E. quoyii are capable of flexible spatial learning and suggest that future studies focus on a range of lizard species which differ in phylogeny and/or ecology, using biologically relevant cognitive tasks, in an effort to bridge the cognitive divide between ecto- and endotherms.     

The spatiotemporal learning rule and its efficiency in separating spatiotemporal patterns

The hippocampus plays an important role in the course of establishing long-term memory, i.e., to make short-term memory of spatially and temporally associated input information. In 1996 (Tsukada et al. 1996), the spatiotemporal learning rule was proposed based on differences observed in hippocampal long-term potentiation (LTP) induced by various spatiotemporal pattern stimuli. One essential point of this learning rule is that the change of synaptic weight depends on both spatial coincidence and the temporal summation of input pulses. We applied this rule to a single-layered neural network and compared its ability to separate spatiotemporal patterns with that of other rules, including the Hebbian learning rule and its extended rules. The simulated results showed that the spatiotemporal learning rule had the highest efficiency in discriminating spatiotemporal pattern sequences, while the Hebbian learning rule (including its extended rules) was sensitive to differences in spatial patterns.     

Reflections on Teaching Social Violence in an Age of Genocide and a Time of War

This article is a reflection by a teacher and a student on "structural-violence pedagogy," the process of teaching and learning about the structures of inequality implicated in various forms of social violence, including those of everyday life, massacre, and genocide. Using a case study of an undergraduate course in anthropology, we explore the complexity and emotionally charged nature of this field, some innovative teaching strategies, and contributions of anthropology to understanding a world marked and marred by war, genocide, racialization, and structural poverty. The teacher shapes a course of study informed by critical pedagogy and theories about violence and power. The student draws on her personal experiences, her intellectual interests, and her yearning for fuller answers to deeper questions as she seeks to reciprocate what is being taught to her. Together, teacher and student explore the promise in the power of teaching for understanding the world as it does exist.     

Sequence length bounds for resolving a deep phylogenetic divergence

In evolutionary biology, genetic sequences carry with them a trace of the underlying tree that describes their evolution from a common ancestral sequence. The question of how many sequence sites are required to recover this evolutionary relationship accurately depends on the model of sequence evolution, the substitution rate, divergence times and the method used to infer phylogenetic history. A particularly challenging problem for phylogenetic methods arises when a rapid divergence event occurred in the distant past. We analyse an idealised form of this problem in which the terminal edges of a symmetric four-taxon tree are some factor (λ) times the length of the interior edge. We determine an order λ² lower bound on the growth rate for the sequence length required to resolve the tree (independent of any particular branch length). We also show that this rate of sequence length growth can be achieved by existing methods (including the simple ‘maximum parsimony’ method), and compare these order λ² bounds with an order λ growth rate for a model that describes low-homoplasy evolution. In the final section, we provide a generic bound on the sequence length requirement for a more general class of Markov processes.     

Biology student teachers’ dialogic talk in inquiry-based instruction

ABSTRACT

Inquiry-based learning has generally accepted by scholars as a most effective teaching approach in biology education. The talk during inquiry-based teaching needs to be practiced. There is less evidence how student teachers talk with students during their inquiry-based biology instruction. This knowledge is needed in supporting student teachers to develop their teachership in biology education. In this qualitative case study, the dialogic talk of biology student teachers (N = 6) was studied in the context of inquiry-based lessons in lower secondary school. The student teachers’ lessons were video and audio recorded and the data was analyzed using content analysis. The student teachers used dialogic talk in their inquiry-based instruction only occasionally, mainly in the examination and the conclusion stages. During the introduction stage, dialogic talk was less used and it was mainly explaining and instructing the content. In the examination stage, student teachers also guided students and stated facts. During the conclusion stage, student teachers mainly explained and also evaluated students’ statements. The lesson’s topics and methods used in inquiry-based learning may enable the dialogic talk of student teacher to some extent. However, teacher education should focus more on scaffolding student teachers’ talk with their students in all kinds of inquiry approaches.