Search

Conventional recommendation systems (RSs) are typically optimized to enhance performance metrics uniformly across all training samples, inadvertently overlooking the needs of diverse user populations. The performance disparity among various populations can harm the model's robustness to sub-populations due to the varying user properties. While large language models (LLMs) show promise in enhancing RS performance, their practical applicability is hindered by high costs, inference latency, and degraded performance on long user queries. To address these challenges, we propose a hybrid task allocation framework designed to promote social good by equitably serving all user groups. By adopting a two-phase approach, we promote a strategic assignment of tasks for efficient and responsible adaptation of LLMs. Our strategy works by first identifying the weak and inactive users that receive a suboptimal ranking performance by RSs. Next, we use an in-context learning approach for such users, wherein each user interaction history is contextualized as a distinct ranking task. We evaluate our hybrid framework by incorporating eight different recommendation algorithms and three different LLMs -- both open and close-sourced. Our results on three real-world datasets show a significant reduction in weak users and improved robustness to subpopulations without disproportionately escalating costs., Comment: arXiv admin note: text overlap with arXiv:2405.00824

In this work, we study the effect of noise-induced coherence on the performance analysis of a degenerate four-level quantum heat engine with particular focus on the universal nature of efficiency, which refers to the appearance of the first two universal terms $\eta_C/2$ and $\eta_C^2/8$ in series expansion of the efficiency at maximum power under a left-right symmetry in the system. Firstly, for a two-parameter optimization scheme, we derive an analytic expression for the efficiency at maximum power for the near-equilibrium condition and show that presence of noise-induced coherence breaks the left-right symmetry of the system. However, when the operation of the engine is restricted to either high-temperature or low-temperature regime, we discuss the conditions under which the left-right symmetry can be retained in each case, giving rise to the universal characteristic of efficiency. In case of one-parameter optimization, we show that while the universality of the first linear term $\eta_c/2$ is robust and holds consistently across all conditions, the universality of the quadratic term $\eta_C^2/8$ depends on the constraints imposed on the control parameters. Finally, we examine the behavior of power as a function of noise-induced coherence parameter highlighting the role of matter-field coupling in determining the suitable operation regime for the heat engine to reap the benefits of noise-induced coherence.

We study a quantum harmonic Otto engine under a hot squeezed thermal reservoir with asymmetry between the two adiabatic branches introduced by considering different speeds of the driving protocols. In the first configuration, the driving protocol for the expansion stroke is sudden-switch in nature and compression stroke is driven adiabatically, while the second configuration deals with the converse situation. In both cases, we obtain analytic expressions for the upper bound on efficiency and efficiency at optimal work output, which reveals a significant difference between the two configurations. Additionally, we find that the maximum achievable efficiency in sudden expansion case is 1/2 only while it approaches unity for the sudden compression stroke. Further, we study the effect of increasing degree of squeezing on the efficiency and work output of the engine and indicate the optimal operational regime for both configurations under consideration. Finally, by studying the full phase-diagram of the Otto cycle we observe that the operational region of the engine mode grows with increasing squeezing at the expense of refrigeration regime., Comment: 9 pages, 6 figures, and 1 table

To reduce the communication overhead caused by parallel training of multiple clients, various federated learning (FL) techniques use random client sampling. Nonetheless, ensuring the efficacy of random sampling and determining the optimal number of clients to sample in federated recommender systems (FRSs) remains challenging due to the isolated nature of each user as a separate client. This challenge is exacerbated in models where public and private features can be separated, and FL allows communication of only public features (item gradients). In this study, we establish sample complexity bounds that dictate the ideal number of clients required for improved communication efficiency and retained accuracy in such models. In line with our theoretical findings, we empirically demonstrate that RS-FairFRS reduces communication cost (~47%). Second, we demonstrate the presence of class imbalance among clients that raises a substantial equity concern for FRSs. Unlike centralized machine learning, clients in FRS can not share raw data, including sensitive attributes. For this, we introduce RS-FairFRS, first fairness under unawareness FRS built upon random sampling based FRS. While random sampling improves communication efficiency, we propose a novel two-phase dual-fair update technique to achieve fairness without revealing protected attributes of active clients participating in training. Our results on real-world datasets and different sensitive features illustrate a significant reduction in demographic bias (~approx40\%), offering a promising path to achieving fairness and communication efficiency in FRSs without compromising the overall accuracy of FRS.

Conventional recommendation systems (RSs) are typically optimized to enhance performance metrics uniformly across all training samples. This makes it hard for data-driven RSs to cater to a diverse set of users due to the varying properties of these users. The performance disparity among various populations can harm the model's robustness with respect to sub-populations. While recent works have shown promising results in adapting large language models (LLMs) for recommendation to address hard samples, long user queries from millions of users can degrade the performance of LLMs and elevate costs, processing times and inference latency. This challenges the practical applicability of LLMs for recommendations. To address this, we propose a hybrid task allocation framework that utilizes the capabilities of both LLMs and traditional RSs. By adopting a two-phase approach to improve robustness to sub-populations, we promote a strategic assignment of tasks for efficient and responsible adaptation of LLMs. Our strategy works by first identifying the weak and inactive users that receive a suboptimal ranking performance by RSs. Next, we use an in-context learning approach for such users, wherein each user interaction history is contextualized as a distinct ranking task and given to an LLM. We test our hybrid framework by incorporating various recommendation algorithms -- collaborative filtering and learning-to-rank recommendation models -- and two LLMs -- both open and close-sourced. Our results on three real-world datasets show a significant reduction in weak users and improved robustness of RSs to sub-populations $(\approx12\%)$ and overall performance without disproportionately escalating costs.

Aperiodic activity is a physiologically distinct component of the electrophysiological power spectrum. It is suggested to reflect the balance of excitation and inhibition in the brain, within selected frequency bands. However, the impact of recurrent seizures on aperiodic activity remains unknown, particularly in patients with severe bilateral seizures. Here, we hypothesised greater aperiodic abnormality in the epileptogenic zone, in patients with focal to bilateral tonic clonic (FBTC) seizures, and earlier age of seizure onset. Pre-operative magnetoencephalography (MEG) recordings were acquired from 36 patients who achieved complete seizure freedom (Engel I outcome) post-surgical resection. A normative whole brain map of the aperiodic exponent was computed by averaging across subjects for each region in the hemisphere contralateral to the side of resection. Selected regions of interest were then tested for abnormality using deviations from the normative map in terms of z-scores. Resection masks drawn from postoperative structural imaging were used as an approximation of the epileptogenic zone. Patients with FBTC seizures had greater abnormality compared to patients with focal onset seizures alone in the resection volume (p=0.003, area under the ROC curve = 0.78 ). Earlier age of seizure onset was correlated with greater abnormality of the aperiodic exponent in the resection volume (correlation coefficient = -0.3, p= 0.04)) as well as the whole cortex (rho = -0.33, p=0.03). The abnormality of the aperiodic exponent did not significantly differ between the resected and non-resected regions of the brain. Abnormalities in aperiodic components relate to important clinical characteristics such as severity and age of seizure onset. This suggests the potential use of the aperiodic band power component as a marker for severity of epilepsy.

We study the optimal performance of an endoreversible quantum dot heat engine, in which the heat transfer between the system and baths is mediated by qubits, operating under the conditions of a trade-off objective function known as maximum efficient power function defined by the product of power and efficiency of the engine. First, we numerically study the optimization of the efficient power function for the engine under consideration. Then, we obtain some analytic results by applying hightemperature limit and compare the performance of the engine at maximum efficient power functionto the engine operating in the maximum power regime. We find that the engine operating at maximum efficient power function produces at least 88.89% of the maximum power output while at the same time reduces the power loss due to entropy production by considerable amount. We conclude by studying the stochastic simulations of the efficiency of the engine in maximum power and maximum efficient power regime. We find that the engine operating at maximum power is subjected to less power fluctuations as compared to the on one operating at maximum efficient power function., Comment: 8 pages and 8 figures