Comparing feedforward neural networks using independent component analysis on hidden units.

Neural networks are widely used for classification and regression tasks, but they do not always perform well, nor explicitly inform us of the rationale for their predictions. In this study we propose a novel method of comparing a pair of different feedforward neural networks, which draws on independent components obtained by independent component analysis (ICA) on the hidden layers of these networks. It can compare different feedforward neural networks even when they have different structures, as well as feedforward neural networks that learned partially different datasets, yielding insights into their functionality or performance. We evaluate the proposed method by conducting three experiments with feedforward neural networks that have one hidden layer, and verify whether a pair of feedforward neural networks can be compared by the proposed method when the numbers of hidden units in the layer are different, when the datasets are partially different, and when activation functions are different. The results show that similar independent components are extracted from two feedforward neural networks, even when the three circumstances above are different. Our experiments also reveal that mere comparison of weights or activations does not lead to identifying similar relationships. Through the extraction of independent components, the proposed method can assess whether the internal processing of one neural network resembles that of another. This approach has the potential to help understand the performance of neural networks.

Causal intuition in the indefinite world: Meta-analysis and simulations.

Modeling our causal intuition can contribute to understanding our behavior. In this paper, we introduce a causal induction model called proportion of assumed-to-be rare instances (pARIs) and examine its adaptive properties. We employ the two-stage theory of causal induction proposed by Hattori and Oaksford in 2007, which divides causal induction into two stages: first, observed events are sifted and likely candidates are extracted; second, each of them is verified through intervention. Here, we focus on the first stage. We conducted a meta-analysis and computer simulations in a similar way to Hattori and Oaksford (2007) but with some corrections and improvements. We added two experiments and excluded one in our reconstructed meta-analysis and augmented the simulations by correcting two problems. Our meta-analysis results show that pARIs outperforms more than 40 existing models in terms of data fit from human causal induction experiments while being simpler. Additionally, our simulation results show that pARIs outperforms DFH in terms of population covariation detection, especially under small sample sizes and rarity of events. Overall, pARIs qualifies as one of the best models for the first stage of causal induction. These findings may enable a deeper understanding of our cognitive biases. The first stage can now be considered a causal discovery stage where the topology of causal models is to be hypothesized.

Conveying Intention by Motions With Awareness of Information Asymmetry.

Humans sometimes attempt to infer an artificial agent's mental state based on mere observations of its behavior. From the agent's perspective, it is important to choose actions with awareness of how its behavior will be considered by humans. Previous studies have proposed computational methods to generate such publicly self-aware motion to allow an agent to convey a certain intention by motions that can lead a human observer to infer what the agent is aiming to do. However, little consideration has been given to the effect of information asymmetry between the agent and a human, or to the gaps in their beliefs due to different observations from their respective perspectives. This paper claims that information asymmetry is a key factor for conveying intentions with motions. To validate the claim, we developed a novel method to generate intention-conveying motions while considering information asymmetry. Our method utilizes a Bayesian public self-awareness model that effectively simulates the inference of an agent's mental states as attributed to the agent by an observer in a partially observable domain. We conducted two experiments to investigate the effects of information asymmetry when conveying intentions with motions by comparing the motions from our method with those generated without considering information asymmetry in a manner similar to previous work. The results demonstrate that by taking information asymmetry into account, an agent can effectively convey its intention to human observers.

Softsatisficing: Risk-sensitive softmax action selection.

Animals, humans, and organizations are known to adjust how (much) they explore complex environments that exceed their information processing capacity, rather than relentlessly search for the optimal action. The adjusted depth of exploration is supposed to depend on the aspiration level internal to the agent. This action selection tendency is known as satisficing. The Risk-sensitive Satisficing (RS) model implements satisficing in the reinforcement learning framework through conversion of action values into gains (or losses) relative to the aspiration level. The risk-sensitive evaluation of action values by RS has been shown to be effective in reinforcement learning. In this paper, first we analyze RS in comparison with UCB and Thompson sampling algorithms. We also show that RS shows differential risk-attitudes considering the risks. Then we propose the Softsatisficing policy that is a stochastic equivalent of RS and further analyze the exploratory behavior of risk-sensitive satisficing that RS and Softsatisficing implement. We emphasize that Softsatisficing has the potential of modeling risk-sensitive foraging and other decision-making behaviors by humans, animals, and organizations.

Effects of salmon cartilage proteoglycan on obesity in mice fed with a high-fat diet.

This study investigated the effects of salmon nasal cartilage proteoglycan (PG), which shows anti-inflammatory properties, on obesity induced by high-fat diet (HFD) in a mouse model. Mice were fed either a HFD or normal diet (ND), with or without PG, for 8-12 weeks. After 12 weeks, the body weight of mice fed with PG-free HFD was 54.08 ± 4.67 g, whereas that of mice fed with HFD containing PG was 41.83 ± 4.97 g. The results suggest that the increase in body weight was attenuated in mice fed with HFD containing PG. This effect was not observed in mice fed with ND. The PG administration suppressed the elevation of serum lipids (the level of serum lipids ranged between 54% and 69% compared to 100% in mice fed with PG-free HFD) and the upregulated mRNA expression of sterol regulatory element-binding protein-1c (SREBP-1c), which is a transcription factor that acts as a master regulator of lipogenic gene expression in the liver (the expression level was 77.5% compared to 100% in mice fed with PG-free HFD). High leptin levels in mice fed with PG-free HFD were observed during fasting (average at 14,376 ng/ml), and they did not increase after refeeding (average of 14,263 ng/ml), whereas serum leptin levels in mice fed with HFD containing PG were low during fasting (average of 6481 ng/ml) and increased after refeeding (average 13,382 ng/ml). These results suggest that PG feeding has an anti-obesity effect and that the regulation of SREBP-1c and leptin secretion play a role in this effect.

The protective effect of Centella asiatica and its constituent, araliadiol on neuronal cell damage and cognitive impairment.

Alzheimer's disease (AD) is characterized by progressive cognitive decline, and the number of affected individuals has increased worldwide. However, there are no effective treatments for AD. Therefore, it is important to prevent the onset of dementia. Oxidative stress and endoplasmic reticulum (ER) stress are increased in the brains of AD patients, and are postulated to induce neuronal cell death and cognitive dysfunction. In this study, Centella asiatica, a traditional Indian medicinal herb, were fractionated and compared for their protective effects against glutamate and tunicamycin damage. Araliadiol was identified as a component from the fraction with the highest activity. Further, murine hippocampal cells (HT22) were damaged by glutamate, an oxidative stress inducer. C. asiatica and araliadiol suppressed cell death and reactive oxygen species production. HT22 cells were also injured by tunicamycin, an ER stress inducer. C. asiatica and araliadiol prevented cell death by mainly inhibiting PERK phosphorylation; additionally, C. asiatica also suppressed the expression levels of GRP94 and BiP. In Y-maze test, oral administration of araliadiol (10 mg/kg/day) for 7 days ameliorated the arm alternation ratio in mice with scopolamine-induced cognitive impairment. These results suggest that C. asiatica and its active component, araliadiol, have neuroprotective effects, which may prevent cognitive dysfunction.

The asymmetric learning rates of murine exploratory behavior in sparse reward environments.

Goal-oriented behaviors of animals can be modeled by reinforcement learning algorithms. Such algorithms predict future outcomes of selected actions utilizing action values and updating those values in response to the positive and negative outcomes. In many models of animal behavior, the action values are updated symmetrically based on a common learning rate, that is, in the same way for both positive and negative outcomes. However, animals in environments with scarce rewards may have uneven learning rates. To investigate the asymmetry in learning rates in reward and non-reward, we analyzed the exploration behavior of mice in five-armed bandit tasks using a Q-learning model with differential learning rates for positive and negative outcomes. The positive learning rate was significantly higher in a scarce reward environment than in a rich reward environment, and conversely, the negative learning rate was significantly lower in the scarce environment. The positive to negative learning rate ratio was about 10 in the scarce environment and about 2 in the rich environment. This result suggests that when the reward probability was low, the mice tend to ignore failures and exploit the rare rewards. Computational modeling analysis revealed that the increased learning rates ratio could cause an overestimation of and perseveration on rare-rewarding events, increasing total reward acquisition in the scarce environment but disadvantaging impartial exploration.

Acorus calamus extract and its component α-asarone attenuate murine hippocampal neuronal cell death induced by l-glutamate and tunicamycin.

The Asian traditional medicinal plant Acorus calamus and its component α-asarone exhibited various biological activities, such as antiinflammation and antioxidant effects. In the present study, we investigated the in vitro effects of A. calamus extract and α-asarone on oxidative stress- and endoplasmic reticulum (ER) stress-induced cell death in hippocampal HT22 cells. A. calamus extract and α-asarone both significantly suppressed cell death induced by the oxidative stress inducer l-glutamate and ER stress inducer tunicamycin. A. calamus extract and α-asarone also significantly reduced reactive oxygen species (ROS) production induced by l-glutamate. Moreover, A. calamus extract and α-asarone suppressed the phosphorylation of protein kinase RNA-like ER kinase (PERK) induced by tunicamycin. These results suggest that A. calamus extract and α-asarone protect hippocampal cells from oxidative stress and ER stress by decreasing ROS production and suppressing PERK signaling, respectively. α-Asarone has potential as a potent therapeutic candidate for neurodegenerative diseases, including Alzheimer's disease.

Self-incremental learning vector quantization with human cognitive biases.

Human beings have adaptively rational cognitive biases for efficiently acquiring concepts from small-sized datasets. With such inductive biases, humans can generalize concepts by learning a small number of samples. By incorporating human cognitive biases into learning vector quantization (LVQ), a prototype-based online machine learning method, we developed self-incremental LVQ (SILVQ) methods that can be easily interpreted. We first describe a method to automatically adjust the learning rate that incorporates human cognitive biases. Second, SILVQ, which self-increases the prototypes based on the method for automatically adjusting the learning rate, is described. The performance levels of the proposed methods are evaluated in experiments employing four real and two artificial datasets. Compared with the original learning vector quantization algorithms, our methods not only effectively remove the need for parameter tuning, but also achieve higher accuracy from learning small numbers of instances. In the cases of larger numbers of instances, SILVQ can still achieve an accuracy that is equal to or better than those of existing representative LVQ algorithms. Furthermore, SILVQ can learn linearly inseparable conceptual structures with the required and sufficient number of prototypes without overfitting.

Exogenous Application of Proteoglycan to the Cell Surface Microenvironment Facilitates to Chondrogenic Differentiation and Maintenance.

Osteoarthritis (OA), a disease that greatly impacts quality of life, has increasing worldwide prevalence as the population ages. However, its pathogenic mechanisms have not been fully elucidated and current therapeutic treatment strategies are inadequate. In recent years, abnormal endochondral ossification in articular cartilage has received attention as a pathophysiological mechanism in OA. Cartilage is composed of abundant extracellular matrix components, which are involved in tissue maintenance and regeneration, but how these factors affect endochondral ossification is not clear. Here, we show that the application of aggrecan-type proteoglycan from salmon nasal cartilage (sPG) exhibited marked proliferative capacity through receptor tyrosine kinases in chondroprogenitor cells, and also exhibited differentiation and three-dimensional structure formation via phosphorylation of Insulin-like Growth Factor-1 Receptor and Growth Differentiation Factor 5 expression. Furthermore, sPG inhibited calcification via expression of Runx2 and Col10 (factors related to induction of calcification), while increasing Mgp, a mineralization inhibitory factor. As a result of analyzing the localization of sPG applied to the cells, it was localized on the surface of the cell membrane. In this study, we found that sPG, as a biomaterial, could regulate cell proliferation, differentiation and calcification inhibition by acting on the cell surface microenvironment. Therefore, sPG may be the foundation for a novel therapeutic approach for cartilage maintenance and for improved symptoms in OA.

A category theoretic approach to metaphor comprehension: Theory of indeterminate natural transformation.

We propose the theory of indeterminate natural transformation (TINT) to investigate the dynamical creation of meaning as an association relationship between images, focusing on metaphor comprehension as an example. TINT models meaning creation as a type of stochastic process based on mathematical structure and defined by association relationships, such as morphisms in category theory, to represent the indeterminate nature of structure-structure interactions between the systems of image meanings. Such interactions are formulated in terms of the so-called coslice categories and functors as structure-preserving correspondences between them. The relationship between such functors is "indeterminate natural transformation," the central notion in TINT, which models the creation of meanings in a precise manner. For instance, metaphor comprehension is modeled by the construction of indeterminate natural transformations from a canonically defined functor, which we call the base-of-metaphor functor.

A new method of Bayesian causal inference in non-stationary environments.

Bayesian inference is the process of narrowing down the hypotheses (causes) to the one that best explains the observational data (effects). To accurately estimate a cause, a considerable amount of data is required to be observed for as long as possible. However, the object of inference is not always constant. In this case, a method such as exponential moving average (EMA) with a discounting rate is used to improve the ability to respond to a sudden change; it is also necessary to increase the discounting rate. That is, a trade-off is established in which the followability is improved by increasing the discounting rate, but the accuracy is reduced. Here, we propose an extended Bayesian inference (EBI), wherein human-like causal inference is incorporated. We show that both the learning and forgetting effects are introduced into Bayesian inference by incorporating the causal inference. We evaluate the estimation performance of the EBI through the learning task of a dynamically changing Gaussian mixture model. In the evaluation, the EBI performance is compared with those of the EMA and a sequential discounting expectation-maximization algorithm. The EBI was shown to modify the trade-off observed in the EMA.

Extended Bayesian inference incorporating symmetry bias.

We start by proposing a causal induction model that incorporates symmetry bias. This model has two parameters that control the strength of symmetry bias and includes conditional probability and conventional models of causal induction as special cases. We calculated the determination coefficients between assessments by participants in eight types of causal induction experiments and the estimated values using the proposed model. The mean coefficient of determination was 0.93. Thus, it can reproduce causal induction of human judgment with high accuracy. We further propose a human-like Bayesian inference method to replace the conditional probability in Bayesian inference with the aforementioned causal induction model. In this method, two components coexist: the component of Bayesian inference, which updates the degree of confidence for each hypothesis, and the component of inverse Bayesian inference that modifies the model of each hypothesis. In other words, this method allows not only inference but also simultaneous learning. Our study demonstrates that the method addresses unsteady situations where the target of inference occasionally changes not only by making inferences based on knowledge (model) and observation data, but also by modifying the model itself.

Guaranteed satisficing and finite regret: Analysis of a cognitive satisficing value function.

As reinforcement learning algorithms are being applied to increasingly complicated and realistic tasks, it is becoming increasingly difficult to solve such problems within a practical time frame. Hence, we focus on a satisficing strategy that looks for an action whose value is above the aspiration level (analogous to the break-even point), rather than the optimal action. In this paper, we introduce a simple mathematical model called risk-sensitive satisficing (RS) that implements a satisficing strategy by integrating risk-averse and risk-prone attitudes under the greedy policy. We apply the proposed model to the K-armed bandit problems, which constitute the most basic class of reinforcement learning tasks, and prove two propositions. The first is that RS is guaranteed to find an action whose value is above the aspiration level. The second is that the regret (expected loss) of RS is upper bounded by a finite value, given that the aspiration level is set to an "optimal level" so that satisficing implies optimizing. We confirm the results through numerical simulations and compare the performance of RS with that of other representative algorithms for the K-armed bandit problems.

The Psychology of Uncertainty and Three-Valued Truth Tables.

Psychological research on people's understanding of natural language connectives has traditionally used truth table tasks, in which participants evaluate the truth or falsity of a compound sentence given the truth or falsity of its components in the framework of propositional logic. One perplexing result concerned the indicative conditional if A then C which was often evaluated as true when A and C are true, false when A is true and C is false but irrelevant" (devoid of value) when A is false (whatever the value of C). This was called the "psychological defective table of the conditional." Here we show that far from being anomalous the "defective" table pattern reveals a coherent semantics for the basic connectives of natural language in a trivalent framework. This was done by establishing participants' truth tables for negation, conjunction, disjunction, conditional, and biconditional, when they were presented with statements that could be certainly true, certainly false, or neither. We review systems of three-valued tables from logic, linguistics, foundations of quantum mechanics, philosophical logic, and artificial intelligence, to see whether one of these systems adequately describes people's interpretations of natural language connectives. We find that de Finetti's (1936/1995) three-valued system is the best approximation to participants' truth tables.

Understanding Conditionals in the East: A Replication Study of With Easterners.

The new probabilistic approaches to the natural language conditional imply that there is a parallel relation between indicative conditionals (ICs) "if s then b" and conditional bets (CBs) "I bet $1 that if s then b" in two aspects. First, the probability of an IC and the probability of winning a CB are both the conditional probability, P(s|b). Second, both an IC and a CB have a third value "void" (neither true nor false, neither wins nor loses) when the antecedent is false (¬s). These aspects of the parallel relation have been found in Western participants. In the present study, we investigated whether this parallel is also present in Eastern participants. We replicated the study of Politzer et al. (2010) with Chinese and Japanese participants and made two predictions. First, Eastern participants will tend to engage in more holistic cognition and take all possible cases, including ¬s, into account when they judge the probability of conditional: Easterners may assess the probability of antecedent s out of all possible cases, P(s), and then may focus on consequent b out of s, P(b|s). Consequently, Easterners may judge the probability of the conditional, and of winning the bet, to be P(s) ∗ P(b|s) = P(s & b), and false/losing the bet as P(s) ∗ P(¬b|s) = P(s & ¬b). Second, Eastern participants will tend to be strongly affected by context, and they may not show parallel relationships between ICs and CBs. The results indicate no cultural differences in judging the false antecedent cases: Eastern participants judged false antecedent cases as not making the IC true nor false and as not being winning or losing outcomes. However, there were cultural differences when asked about the probability of a conditional. Consistent with our hypothesis, Eastern participants had a greater tendency to take all possible cases into account, especially in CBs. We discuss whether these results can be explained by a hypothesized tendency for Eastern people to think in more holistic and context-dependent terms than Western people.

Evaluation of the effect of salmon nasal proteoglycan on biomarkers for cartilage metabolism in individuals with knee joint discomfort: A randomized double-blind placebo-controlled clinical study.

A randomized double-blind placebo-controlled clinical trial was conducted to evaluate the chondroprotective action of salmon nasal cartilage proteoglycan on joint health. The effect of oral administration of proteoglycan (10 mg/day) on cartilage metabolism was evaluated in individuals with knee joint discomfort but without diagnosis of knee osteoarthritis. The average age of patients was 52.6±1.1 years old. The effect of proteoglycan was evaluated by analyzing markers for type II collagen degradation (C1,2C) and synthesis (PIICP), and the ratio of type II collagen degradation to synthesis. The results indicated that the change in C1,2C levels significantly differed in the proteoglycan group compared with the placebo group following 16 weeks intervention among subjects with high levels of knee pain and physical dysfunction (total score of Japan Knee Osteoarthritis Measure ≥41) and subjects with constant knee pain (both P<0.05). There was a greater increase in PIICP levels in the proteoglycan group than the placebo group following intervention, although this difference was not significant in both sets of patients. Thus, the C1,2C/PIICP ratios decreased in the proteoglycan group, whereas they slightly increased in the placebo group following the intervention. Furthermore, no test supplement-related adverse events were observed during the intervention. Therefore, oral administration of salmon nasal cartilage proteoglycan at a dose of 10 mg/day may exert a chondroprotective action in subjects with knee joint discomfort. This effect was achieved by improving cartilage metabolism (reducing type II collagen degradation and enhancing type II collagen synthesis), without causing apparent adverse effects.

Robotic action acquisition with cognitive biases in coarse-grained state space.

Some of the authors have previously proposed a cognitively inspired reinforcement learning architecture (LS-Q) that mimics cognitive biases in humans. LS-Q adaptively learns under uniform, coarse-grained state division and performs well without parameter tuning in a giant-swing robot task. However, these results were shown only in simulations. In this study, we test the validity of the LS-Q implemented in a robot in a real environment. In addition, we analyze the learning process to elucidate the mechanism by which the LS-Q adaptively learns under the partially observable environment. We argue that the LS-Q may be a versatile reinforcement learning architecture, which is, despite its simplicity, easily applicable and does not require well-prepared settings.

Effects of Salmon Nasal Cartilage Proteoglycan on Plasma Glucose Concentration and Active Glucose Transport in the Small Intestine.

Recently, proteoglycan was purified from the nasal cartilage of salmon. Although several physiological effects have been reported, the effect of salmon nasal cartilage proteoglycan (salmon PG) on glucose metabolism remains unclear. We studied the effect of salmon PG on rat plasma glucose levels. Oral administration of 1% salmon PG significantly attenuated the increase in portal plasma glucose levels following an oral glucose tolerance test (OGTT). Additionally 1% salmon PG delayed the increase in peripheral glucose concentration induced by the OGTT. Mucosal administration of 1% salmon PG significantly decreased active glucose transport using the everted jejunal sac method. Furthermore, transmural potential difference (ΔPD) measurements using the everted jejunum revealed that 1% salmon PG significantly decreased glucose-dependent and phlorhizin (inhibitor of sodium-glucose co-transporter 1; SGLT1)-sensitive ΔPD. These results suggest that salmon PG decreases glucose absorption via SGLT1 in the jejunum, thereby attenuating the increase in portal and peripheral plasma glucose levels in rats.

Cognitively inspired reinforcement learning architecture and its application to giant-swing motion control.

Many algorithms and methods in artificial intelligence or machine learning were inspired by human cognition. As a mechanism to handle the exploration-exploitation dilemma in reinforcement learning, the loosely symmetric (LS) value function that models causal intuition of humans was proposed (Shinohara et al., 2007). While LS shows the highest correlation with causal induction by humans, it has been reported that it effectively works in multi-armed bandit problems that form the simplest class of tasks representing the dilemma. However, the scope of application of LS was limited to the reinforcement learning problems that have K actions with only one state (K-armed bandit problems). This study proposes LS-Q learning architecture that can deal with general reinforcement learning tasks with multiple states and delayed reward. We tested the learning performance of the new architecture in giant-swing robot motion learning, where uncertainty and unknown-ness of the environment is huge. In the test, the help of ready-made internal models or functional approximation of the state space were not given. The simulations showed that while the ordinary Q-learning agent does not reach giant-swing motion because of stagnant loops (local optima with low rewards), LS-Q escapes such loops and acquires giant-swing. It is confirmed that the smaller number of states is, in other words, the more coarse-grained the division of states and the more incomplete the state observation is, the better LS-Q performs in comparison with Q-learning. We also showed that the high performance of LS-Q depends comparatively little on parameter tuning and learning time. This suggests that the proposed method inspired by human cognition works adaptively in real environments.