Information-Theoretic Representation Learning for Positive-Unlabeled Classification

Recent advances in weakly supervised classification allow us to train a classifier from only positive and unlabeled (PU) data. However, existing PU classification methods typically require an accurate estimate of the class-prior probability, a critical bottleneck particularly for high-dimensional data. This problem has been commonly addressed by applying principal component analysis in advance, but such unsupervised dimension reduction can collapse the underlying class structure. In this letter, we propose a novel representation learning method from PU data based on the information-maximization principle. Our method does not require class-prior estimation and thus can be used as a preprocessing method for PU classification. Through experiments, we demonstrate that our method, combined with deep neural networks, highly improves the accuracy of PU class-prior estimation, leading to state-of-the-art PU classification performance.

Adapted or Adaptable: How to Manage Entropy Production?

Adaptable or adapted? Whether it is a question of physical, biological, or even economic systems, this problem arises when all these systems are the location of matter and energy conversion. To this interdisciplinary question, we propose a theoretical framework based on the two principles of thermodynamics. Considering a finite time linear thermodynamic approach, we show that non-equilibrium systems operating in a quasi-static regime are quite deterministic as long as boundary conditions are correctly defined. The Novikov–Curzon–Ahlborn derivation applied to non-endoreversible systems then makes it possible to precisely determine the conditions for obtaining characteristic operating points. As a result, power maximization principle (MPP), entropy minimization principle (mEP), efficiency maximization, or waste minimization states are only specific modalities of system operation. We show that boundary conditions play a major role in defining operating points because they define the intensity of the feedback that ultimately characterizes the operation. Armed with these thermodynamic foundations, we show that the intrinsically most efficient systems are also the most constrained in terms of controlling the entropy and dissipation production. In particular, we show that the best figure of merit necessarily leads to a vanishing production of power. On the other hand, a class of systems emerges, which, although they do not offer extreme efficiency or power, have a wide range of use and therefore marked robustness. It therefore appears that the number of degrees of freedom of the system leads to an optimization of the allocation of entropy production.

Adapted or Adaptable: How to Manage Entropy Production?

Adaptable or adapted? Whether it is a question of physical, biological or even economic systems, this problem arises when all these systems are the location of matter and energy conversion. To this interdisciplinary question we propose a theoretical framework based on the two principles of thermodynamics. Considering a finite time linear thermodynamic approach, we show that non-equilibrium systems operating in quasi-static regime are quite deterministic as long as boundary conditions are correctly defined. The Novikov-Curzon-Ahlborn approach [1,2] applied to non-endoreversible systems then makes it possible to precisely determine the conditions for obtaining characteristic operating points. As a result, power maximization principle (MPP), entropy minimization principle(mEP), efficiency maximization, or waste minimization states are only specific modalities of system operation. We show that boundary conditions play a major role in defining operating points because they define the intensity of the feedback that ultimately characterizes the operation. Armed with these thermodynamic foundations, we show that the intrinsically most efficient systems are also the most constrained in terms of controlling the entropy and dissipation production. In particular, we show that the best figure of merit necessarily leads to a vanishing production of power. On the other hand, a class of systems emerges which, although they do not offer extreme efficiency or power, have a wide range of use and therefore marked robustness. It therefore appears that the number of degrees of freedom of the system leads to an optimization of the allocation of entropy production.

[378]Utilitarianism and the Chimera of Public Utility

This chapter begins with a discussion of the relation of Bentham’s utilitarianism to natural law. It argues that Bentham’s break with the past is not as complete as he himself believed. Still, it would be wrong to lump utilitarianism together with natural law. Bentham’s criticism of the sense of morality and justice, and his claim that an action should be judged according to its real effects, are steps towards a realistic theory of permanent value. The chapter then discusses the dissonance between the maximization principle and people’s actual deliberative actions, and the chimera of public utility. It argues that in recent years, it has become quite customary to speak of public utility, of the needs of society and the like, instead of the sum of the pleasure of individuals. However, doing so conceals the fundamental flaws of the utilitarian principle without changing anything with respect to the real issue at hand.

Formalism of a harmonic oscillator in the future-included complex action theory

In a special representation of complex action theory that we call “future-included,” we study a harmonic oscillator model defined with a non-normal Hamiltonian $\hat{H}$, in which a mass $m$ and an angular frequency $\omega$ are taken to be complex numbers. In order for the model to be sensible some restrictions on $m$ and $\omega$ are required. We draw a phase diagram in the plane of the arguments of $m$ and $\omega$, according to which the model is classified into several types. In addition, we formulate two pairs of annihilation and creation operators, two series of eigenstates of the Hamiltonians $\hat{H}$ and $\hat{H}^\dagger$, and coherent states. They are normalized in a modified inner product $I_Q$, with respect to which the Hamiltonian $\hat{H}$ becomes normal. Furthermore, applying to the model the maximization principle that we previously proposed, we obtain an effective theory described by a Hamiltonian that is $Q$-Hermitian, i.e. Hermitian with respect to the modified inner product $I_Q$. The generic solution to the model is found to be the “ground” state. Finally we discuss what the solution implies.

How Much Is Enough in a Perfect World? Cultural Variation in Ideal Levels of Happiness, Pleasure, Freedom, Health, Self-Esteem, Longevity, and Intelligence

The maximization principle—that people aspire to the highest possible level of something good if all practical constraints are removed—is a common yet untested assumption about human nature. We predict that in holistic cultures—where contradiction, change, and context are emphasized—ideal states of being for the self will be more moderate than in other cultures. In two studies ( Ns = 2,392 and 6,239), we asked this question: If participants could choose their ideal level of happiness, pleasure, freedom, health, self-esteem, longevity, and intelligence, what level would they choose? Consistent with predictions, results showed that maximization was less pronounced in holistic cultures; members of holistic cultures aspired to less happiness, pleasure, freedom, health, self-esteem, longevity, and IQ than did members of other cultures. In contrast, no differences emerged on ideals for society. The studies show that the maximization principle is not a universal aspect of human nature and that there are predictable cultural differences in people’s notions of perfection.