Search

In solving the Brans-Dicke (BD) equations in the BD theory of gravity, their linear independence is important. This is due to fact that in solving these equations in cosmology, if the number of unknown quantities is equal to the number of independent equations, then the unknowns can be uniquely determined. In the BD theory, the tensor field $g_{\mu \nu}$ and the BD scalar field $\varphi$ are not two separate fields, but they are coupled together. The reason behind this is a corollary that proposed by V. B. Johri and D. Kalyani in cosmology, which states that the cosmic scale factor of the universe, $a$, and the BD scalar field $\varphi$ are related by a power law. Therefore, when the principle of least action is used to derive the BD equations, the variations $\delta g^{\mu \nu}$ and $\delta \varphi$ should not be considered as two independent dynamical variables. So, there is a constraint on $\delta g^{\mu \nu}$ and $\delta \varphi$ that causes the number of independent BD equations to decrease by one unit, in such a way that in the equations that have been known as BD equations, one of them is redundant. In this paper, we prove this issue, that is, we show that one of these equations, which we choose as the modified Klein-Gordon equation, is not an independent equation, but a result establishing other BD equations, the law of conservation of energy-momentum of matter and Bianchi's identity. Therefore, we should not look at the modified Klein-Gordon equation as an independent field equation in the BD theory, but rather it is included in the other BD equations and should not be mentioned separately as one of the BD equations once again., Comment: 5 pages, formula (5) and Refs. [21-23] added

The premise of semi-supervised learning (SSL) is that combining labeled and unlabeled data yields significantly more accurate models. Despite empirical successes, the theoretical understanding of SSL is still far from complete. In this work, we study SSL for high dimensional sparse Gaussian classification. To construct an accurate classifier a key task is feature selection, detecting the few variables that separate the two classes. % For this SSL setting, we analyze information theoretic lower bounds for accurate feature selection as well as computational lower bounds, assuming the low-degree likelihood hardness conjecture. % Our key contribution is the identification of a regime in the problem parameters (dimension, sparsity, number of labeled and unlabeled samples) where SSL is guaranteed to be advantageous for classification. Specifically, there is a regime where it is possible to construct in polynomial time an accurate SSL classifier. However, % any computationally efficient supervised or unsupervised learning schemes, that separately use only the labeled or unlabeled data would fail. Our work highlights the provable benefits of combining labeled and unlabeled data for {classification and} feature selection in high dimensions. We present simulations that complement our theoretical analysis.

Cr\'epey, Frikha, and Louzi (2023) introduced a multilevel stochastic approximation scheme to compute the value-at-risk of a financial loss that is only simulatable by Monte Carlo. The optimal complexity of the scheme is in $O({\varepsilon}^{-5/2})$, ${\varepsilon} > 0$ being a prescribed accuracy, which is suboptimal when compared to the canonical multilevel Monte Carlo performance. This suboptimality stems from the discontinuity of the Heaviside function involved in the biased stochastic gradient that is recursively evaluated to derive the value-at-risk. To mitigate this issue, this paper proposes and analyzes a multilevel stochastic approximation algorithm that adaptively selects the number of inner samples at each level, and proves that its optimal complexity is in $O({\varepsilon}^{-2}|\ln {\varepsilon}|^{5/2})$. Our theoretical analysis is exemplified through numerical experiments., Comment: 43 pages, 6 tables, 4 figures

In this short article, using a left-invariant Randers metric $F$, we define a new left-invariant Randers metric $\tilde{F}$. We show that $F$ is of Berwald (Douglas) type if and only if $\tilde{F}$ is of Berwald (Douglas) type. In the case of Berwaldian metrics, we give the relation between their flag curvatures. Also, we have studied the relations between their base Riemannian metrics. Finally, as examples, the results are studied in the Heisenberg group and almost Abelian Lie groups.

To better align Large Language Models (LLMs) with human judgment, Reinforcement Learning from Human Feedback (RLHF) learns a reward model and then optimizes it using regularized RL. Recently, direct alignment methods were introduced to learn such a fine-tuned model directly from a preference dataset without computing a proxy reward function. These methods are built upon contrastive losses involving the log-likelihood of (dis)preferred completions according to the trained model. However, completions have various lengths, and the log-likelihood is not length-invariant. On the other side, the cross-entropy loss used in supervised training is length-invariant, as batches are typically averaged token-wise. To reconcile these approaches, we introduce a principled approach for making direct alignment length-invariant. Formally, we introduce a new averaging operator, to be composed with the optimality operator giving the best policy for the underlying RL problem. It translates into averaging the log-likelihood within the loss. We empirically study the effect of such averaging, observing a trade-off between the length of generations and their scores.

Reinforcement Learning (RL) has been used to finetune Large Language Models (LLMs) using a reward model trained from preference data, to better align with human judgment. The recently introduced direct alignment methods, which are often simpler, more stable, and computationally lighter, can more directly achieve this. However, these approaches cannot optimize arbitrary rewards, and the preference-based ones are not the only rewards of interest for LLMs (eg., unit tests for code generation or textual entailment for summarization, among others). RL-finetuning is usually done with a variation of policy gradient, which calls for on-policy or near-on-policy samples, requiring costly generations. We introduce Contrastive Policy Gradient, or CoPG, a simple and mathematically principled new RL algorithm that can estimate the optimal policy even from off-policy data. It can be seen as an off-policy policy gradient approach that does not rely on important sampling techniques and highlights the importance of using (the right) state baseline. We show this approach to generalize the direct alignment method IPO (identity preference optimization) and classic policy gradient. We experiment with the proposed CoPG on a toy bandit problem to illustrate its properties, as well as for finetuning LLMs on a summarization task, using a learned reward function considered as ground truth for the purpose of the experiments.

Both online and offline RLHF methods such as PPO and DPO have been extremely successful in aligning AI with human preferences. Despite their success, the existing methods suffer from a fundamental problem that their optimal solution is highly task-dependent (i.e., not robust to out-of-distribution (OOD) tasks). Here we address this challenge by proposing Self-Improving Robust Preference Optimization SRPO, a practical and mathematically principled offline RLHF framework that is completely robust to the changes in the task. The key idea of SRPO is to cast the problem of learning from human preferences as a self-improvement process, which can be mathematically expressed in terms of a min-max objective that aims at joint optimization of self-improvement policy and the generative policy in an adversarial fashion. The solution for this optimization problem is independent of the training task and thus it is robust to its changes. We then show that this objective can be re-expressed in the form of a non-adversarial offline loss which can be optimized using standard supervised optimization techniques at scale without any need for reward model and online inference. We show the effectiveness of SRPO in terms of AI Win-Rate (WR) against human (GOLD) completions. In particular, when SRPO is evaluated on the OOD XSUM dataset, it outperforms the celebrated DPO by a clear margin of 15% after 5 self-revisions, achieving WR of 90%.

The dominant framework for alignment of large language models (LLM), whether through reinforcement learning from human feedback or direct preference optimisation, is to learn from preference data. This involves building datasets where each element is a quadruplet composed of a prompt, two independent responses (completions of the prompt) and a human preference between the two independent responses, yielding a preferred and a dis-preferred response. Such data is typically scarce and expensive to collect. On the other hand, \emph{single-trajectory} datasets where each element is a triplet composed of a prompt, a response and a human feedback is naturally more abundant. The canonical element of such datasets is for instance an LLM's response to a user's prompt followed by a user's feedback such as a thumbs-up/down. Consequently, in this work, we propose DRO, or \emph{Direct Reward Optimisation}, as a framework and associated algorithms that do not require pairwise preferences. DRO uses a simple mean-squared objective that can be implemented in various ways. We validate our findings empirically, using T5 encoder-decoder language models, and show DRO's performance over selected baselines such as Kahneman-Tversky Optimization (KTO). Thus, we confirm that DRO is a simple and empirically compelling method for single-trajectory policy optimisation.

Despite their ever more widespread deployment throughout society, machine learning algorithms remain critically vulnerable to being spoofed by subtle adversarial tampering with their input data. The prospect of near-term quantum computers being capable of running {quantum machine learning} (QML) algorithms has therefore generated intense interest in their adversarial vulnerability. Here we show that quantum properties of QML algorithms can confer fundamental protections against such attacks, in certain scenarios guaranteeing robustness against classically-armed adversaries. We leverage tools from many-body physics to identify the quantum sources of this protection. Our results offer a theoretical underpinning of recent evidence which suggest quantum advantages in the search for adversarial robustness. In particular, we prove that quantum classifiers are: (i) protected against weak perturbations of data drawn from the trained distribution, (ii) protected against local attacks if they are insufficiently scrambling, and (iii) protected against universal adversarial attacks if they are sufficiently quantum chaotic. Our analytic results are supported by numerical evidence demonstrating the applicability of our theorems and the resulting robustness of a quantum classifier in practice. This line of inquiry constitutes a concrete pathway to advantage in QML, orthogonal to the usually sought improvements in model speed or accuracy., Comment: 9+12 pages, 3 figures. Comments welcome

In this paper we consider an inflating universe with long straight cosmic string along z-axis. We show that the effect of cosmic string can be taken as a perturbation on the background of FRW metric. Then by doing cosmological perturbations on this inflating cosmic string background, we find linearized Einstein field equations. We show that at leading order (ignoring the mixing terms of cosmic string perturbations with gravitational tensor perturbations), the cosmic string appears as an inhomogeneous term on the right hand side of wave equation of tensor perturbations . Then by finding analytical solution of the wave equation for slow-roll inflation, we show how it effects on the spectrum of primordial gravitational waves., Comment: 10 pages, 1 fig

Phase slips occur across all Josephson junctions (JJs) at a rate that increases with the impedance of the junction. In superconducting qubits composed of JJ-array superinductors -- such as fluxonium -- phase slips in the array can lead to decoherence. In particular, phase-slip processes at the individual array junctions can coherently interfere, each with an Aharonov--Casher phase that depends on the offset charges of the array islands. These coherent quantum phase slips (CQPS) perturbatively modify the qubit frequency, and therefore charge noise on the array islands will lead to dephasing. By varying the impedance of the array junctions, we design a set of fluxonium qubits in which the expected phase-slip rate within the JJ-array changes by several orders of magnitude. We characterize the coherence times of these qubits and demonstrate that the scaling of CQPS-induced dephasing rates agrees with our theoretical model. Furthermore, we perform noise spectroscopy of two qubits in regimes dominated by either CQPS or flux noise. We find the noise power spectrum associated with CQPS dephasing appears to be featureless at low frequencies and not $1/f$. Numerical simulations indicate this behavior is consistent with charge noise generated by charge-parity fluctuations within the array. Our findings broadly inform JJ-array-design tradeoffs, relevant for the numerous superconducting qubit designs employing JJ-array superinductors.

We consider magnetic graphene quantum dots (MGQDs) and study the impact of the Aharonov-Bohm (AB) flux and gap on the scattering process of electrons. Our emphasis is on the finite lifetimes of quasi-bound states arising from the interaction between electrons and the magnetic field within the dot. Initially, we calculate the scattering coefficients, scattering efficiency, and probability density by ensuring the continuity of eigenspinors at the boundary of MGQD. The results indicate that as the gap increases, the quasi-bound states reach higher maxima. We show that an increase in AB-flux leads to a generation of quasi-bound states requiring less magnetic field, and the scattering efficiency starts to take non-zero values at smaller MGQD sizes. The analysis of probability density shows that the quasi-bound states, corresponding to non-resonantly excited scattering modes, exhibit a significant improvement in the concentrated density at MGQD. The improvement is a result of reducing the diffraction phenomenon and suppressing the Klein effect through an increase in AB-flux and gap. This increases the probability of retaining the electron for a longer period of time., Comment: 11 pages, 7 figures

Reliable angler behavior data is important for effective fisheries management. Traditionally, such data is gathered through surveys, but an innovative cost-effective approach involves utilizing online platforms and smartphone applications. Previous studies have identified correlations between citizen-reported data from these applications and conventional survey information. However, it remains unclear whether conventional survey data is directly related to citizen-reported data or mainly derived from "intermediate" variables. We applied Bayesian networks to data from conventional surveys, the Angler's Atlas website, the MyCatch smartphone application, and environmental data across Alberta and Ontario, Canada, to detect probabilistic dependencies. Using Bayesian model averaging, we measured the strength of connections between variables. In Ontario, aerial boat counts were directly related to waterbody webpage views, with a 51% probability. In Alberta, creel survey-reported catch rate was directly related to citizen-reported catch rate and fishing duration, though with low probabilities (12% and 6%, respectively). Daily fishing durations were indirectly related, with air temperature and solar radiation as intermediates. These findings suggest that citizen reports can complement traditional methods for evaluating angler behavior., Comment: 18 pages, 2 tables, and 4 figures

We study the effects of a magnetic flux and an inhomogeneous gap on the energy spectrum of graphene magnetic quantum dots (GMQDs). By considering the Dirac equation in the infinite mass framework, we can analytically obtain eigenspinor expressions. By applying boundary conditions, we obtain an energy spectrum equation in terms of system parameters such as radius, magnetic field, energy, flux, and gap. In the infinite limit, we recover Landau levels for graphene in a magnetic field. We show that the energy spectrum increases significantly in the presence of flux and a gap inside the GMQDs, which prolongs the lifetime of the trapped electron states. We show that higher flux also produces new Landau levels of negative angular momentum. Meanwhile, we find that the gap increases the separation between the electron and hole energy bands. As shown in the radial probability analysis, flux and gap emerge as influential factors in controlling electron mobility, affecting confinement, and prolonging the presence of quasi-bound states., Comment: 10 pages, 7 figures. Clarifications and references added

We proceed to obtain an exact analytical solution of the Brans-Dicke (BD) equations for the spatially flat ($k=0$) Friedmann-Lamaitre-Robertson-Walker (FLRW) cosmological model in both cases of the absence and presence of the cosmological constant. The solution method that we use to solve the field equations of the BD equations is called the "invariants of symmetry groups method" (ISG-method). This method is based on the extended Prelle-Singer (PS) method and it employs the Lie point symmetry, $\lambda$-symmetry, and Darboux polynomials (DPs). Indeed, the ISG-method tries to provide two independent first-order invariants associated to the one-parameter Lie groups of transformations keeping ordinary differential equations (ODEs) invariant, as solutions. It should be noted for integrable ODEs, the ISG-method guarantees the extraction of these two invariants. In this work for the BD equations in FLRW cosmological model, we find the Lie point symmetries, $\lambda$-symmetries and DPs, and obtain the basic quantities of the extended PS method (which are the null forms and the integrating factors). By making use of the extended PS method we find two independent first-order invariants, in such a way appropriate cosmological solutions from solving these invariants as a system of algebraic equations are simultaneously obtained. These solutions are wealthy so that they include many known special solutions, such as O'Hanlon-Tupper vacuum solutions, Nariai's solutions, Brans-Dicke dust solutions, inflationary solutions, and etc., Comment: 27 pages, References added, Accepted for publication by IJGMMP

In this article, we employ mathematical concepts as a tool to examine the phenomenon of consciousness experience and logical phenomena. Through our investigation, we aim to demonstrate that our experiences, while not confined to limitations, cannot be neatly encapsulated within a singular collection. Our conscious experience emerges as a result of the developmental and augmentative trajectory of our cognitive system. As our cognitive abilities undergo refinement and advancement, our capacity for logical thinking likewise evolves, thereby manifesting a heightened level of conscious experience. The primary objective of this article is to embark upon a profound exploration of the concept of logical experience, delving into the intricate process by which these experiences are derived from our mind.

We investigate the energy levels of fermions within a circular graphene quantum dot (GQD) subjected to external magnetic and Aharonov-Bohm fields. Solving the eigenvalue equation for two distinct regions allows us to determine the eigenspinors for the valleys $K$ and $K^\prime$. By establishing the continuity of eigenspinors at the GQD interface, we derive an equation that reveals the reliance of energy levels on external physical parameters. Our observations suggest that the symmetry of energy levels hinges on the selected physical parameters. We observe that at low magnetic fields, the energy levels display degeneracy, which diminishes as the field strength increases, coinciding with the convergence of energy levels toward the Landau levels. We illustrate that the introduction of a magnetic flux into the GQD leads to the creation of an energy gap, extending the trapping time of electrons without perturbing the system. Conversely, the addition of gap energy widens the band gap, disrupting the system's symmetry by introducing new energy levels., Comment: 9 pages, 7 figures

Embeddings are a basic initial feature extraction step in many machine learning models, particularly in natural language processing. An embedding attempts to map data tokens to a low-dimensional space where similar tokens are mapped to vectors that are close to one another by some metric in the embedding space. A basic question is how well can such embedding be learned? To study this problem, we consider a simple probability model for discrete data where there is some "true" but unknown embedding where the correlation of random variables is related to the similarity of the embeddings. Under this model, it is shown that the embeddings can be learned by a variant of low-rank approximate message passing (AMP) method. The AMP approach enables precise predictions of the accuracy of the estimation in certain high-dimensional limits. In particular, the methodology provides insight on the relations of key parameters such as the number of samples per value, the frequency of the terms, and the strength of the embedding correlation on the probability distribution. Our theoretical findings are validated by simulations on both synthetic data and real text data., Comment: 12 pages, 7 figures

Internet-of-Things (IoT) is rapidly growing in wireless technology, aiming to connect vast numbers of devices to gather and distribute vital information. Despite individual devices having low energy consumption, the cumulative demand results in significant energy usage. Consequently, the concept of ultra-low-power tags gains appeal. Such tags communicate by reflecting rather than generating the radio frequency (RF) signals by themselves. Thus, these backscatter tags can be low-cost and battery-free. The RF signals can be ambient sources such as wireless-fidelity (Wi-Fi), cellular, or television (TV) signals, or the system can generate them externally. Backscatter channel characteristics are different from conventional point-to-point or cooperative relay channels. These systems are also affected by a strong interference link between the RF source and the tag besides the direct and backscattering links, making signal detection challenging. This paper provides an overview of the fundamentals, challenges, and ongoing research in signal detection for AmBC networks. It delves into various detection methods, discussing their advantages and drawbacks. The paper's emphasis on signal detection sets it apart and positions it as a valuable resource for IoT and wireless communication professionals and researchers., Comment: Accepted for publication in the IEEE Access

Reinforcement learning from human feedback (RLHF) has emerged as the main paradigm for aligning large language models (LLMs) with human preferences. Typically, RLHF involves the initial step of learning a reward model from human feedback, often expressed as preferences between pairs of text generations produced by a pre-trained LLM. Subsequently, the LLM's policy is fine-tuned by optimizing it to maximize the reward model through a reinforcement learning algorithm. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In this study, we introduce an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF). In the context of a tabular policy representation, we present a novel algorithmic solution, Nash-MD, founded on the principles of mirror descent. This algorithm produces a sequence of policies, with the last iteration converging to the regularized Nash equilibrium. Additionally, we explore parametric representations of policies and introduce gradient descent algorithms for deep-learning architectures. To demonstrate the effectiveness of our approach, we present experimental results involving the fine-tuning of a LLM for a text summarization task. We believe NLHF offers a compelling avenue for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences.