Search

Reviews and ratings by users form a central component in several widely used products today (e.g., product reviews, ratings of online content, etc.), but today's platforms for managing such reviews are ad-hoc and vulnerable to various forms of tampering and hijack by fake reviews either by bots or motivated paid workers. We define a new metric called 'hijack-resistance' for such review platforms, and then present TrustRate, an end-to-end decentralized, hijack-resistant platform for authentic, anonymous, tamper-proof reviews. With a prototype implementation and evaluation at the scale of thousands of nodes, we demonstrate the efficacy and performance of our platform, towards a new paradigm for building products based on trusted reviews by end users without having to trust a single organization that manages the reviews., Comment: 23 pages. Poster at The 24th Privacy Enhancing Technologies Symposium, 2024, Bristol, United Kingdom

The advent of autonomous vehicles has heralded a transformative era in transportation, reshaping the landscape of mobility through cutting-edge technologies. Central to this evolution is the integration of Artificial Intelligence (AI) and learning algorithms, propelling vehicles into realms of unprecedented autonomy. This paper provides a comprehensive exploration of the evolutionary trajectory of AI within autonomous vehicles, tracing the journey from foundational principles to the most recent advancements. Commencing with a current landscape overview, the paper delves into the fundamental role of AI in shaping the autonomous decision-making capabilities of vehicles. It elucidates the steps involved in the AI-powered development life cycle in vehicles, addressing ethical considerations and bias in AI-driven software development for autonomous vehicles. The study presents statistical insights into the usage and types of AI/learning algorithms over the years, showcasing the evolving research landscape within the automotive industry. Furthermore, the paper highlights the pivotal role of parameters in refining algorithms for both trucks and cars, facilitating vehicles to adapt, learn, and improve performance over time. It concludes by outlining different levels of autonomy, elucidating the nuanced usage of AI and learning algorithms, and automating key tasks at each level. Additionally, the document discusses the variation in software package sizes across different autonomy levels, Comment: 13 pages

We introduce a formal statistical definition for the problem of backdoor detection in machine learning systems and use it to analyze the feasibility of such problems, providing evidence for the utility and applicability of our definition. The main contributions of this work are an impossibility result and an achievability result for backdoor detection. We show a no-free-lunch theorem, proving that universal (adversary-unaware) backdoor detection is impossible, except for very small alphabet sizes. Thus, we argue, that backdoor detection methods need to be either explicitly, or implicitly adversary-aware. However, our work does not imply that backdoor detection cannot work in specific scenarios, as evidenced by successful backdoor detection methods in the scientific literature. Furthermore, we connect our definition to the probably approximately correct (PAC) learnability of the out-of-distribution detection problem.

In the realm of recommender systems, the ubiquitous adoption of deep neural networks has emerged as a dominant paradigm for modeling diverse business objectives. As user bases continue to expand, the necessity of personalization and frequent model updates have assumed paramount significance to ensure the delivery of relevant and refreshed experiences to a diverse array of members. In this work, we introduce an innovative meta-learning solution tailored to the personalization of models for individual members and other entities, coupled with the frequent updates based on the latest user interaction signals. Specifically, we leverage the Model-Agnostic Meta Learning (MAML) algorithm to adapt per-task sub-networks using recent user interaction data. Given the near infeasibility of productionizing original MAML-based models in online recommendation systems, we propose an efficient strategy to operationalize meta-learned sub-networks in production, which involves transforming them into fixed-sized vectors, termed meta embeddings, thereby enabling the seamless deployment of models with hundreds of billions of parameters for online serving. Through extensive experimentation on production data drawn from various applications at LinkedIn, we demonstrate that the proposed solution consistently outperforms the baseline models of those applications, including strong baselines such as using wide-and-deep ID based personalization approach. Our approach has enabled the deployment of a range of highly personalized AI models across diverse LinkedIn applications, leading to substantial improvements in business metrics as well as refreshed experience for our members.

The availability of large annotated data can be a critical bottleneck in training machine learning algorithms successfully, especially when applied to diverse domains. Weak supervision offers a promising alternative by accelerating the creation of labeled training data using domain-specific rules. However, it requires users to write a diverse set of high-quality rules to assign labels to the unlabeled data. Automatic Rule Induction (ARI) approaches circumvent this problem by automatically creating rules from features on a small labeled set and filtering a final set of rules from them. In the ARI approach, the crucial step is to filter out a set of a high-quality useful subset of rules from the large set of automatically created rules. In this paper, we propose an algorithm (Filtering of Automatically Induced Rules) to filter rules from a large number of automatically induced rules using submodular objective functions that account for the collective precision, coverage, and conflicts of the rule set. We experiment with three ARI approaches and five text classification datasets to validate the superior performance of our algorithm with respect to several semi-supervised label aggregation approaches. Further, we show that achieves statistically significant results in comparison to existing rule-filtering approaches., Comment: EACL 2024

The role of self-consistency in Bogoliubov-de Gennes equations is frequently underestimated in the investigation of in-gap states created by magnetic impurities in s-wave superconductors. Our research focuses on the impact of self-consistency on the in-gap states produced by magnetic stuctures on superconductors, specifically evaluating the density of states, the in-gap bands, and their topological attributes. Here, we show results ranging from single impurity to finite chains, and infinite ferromagnetic spin chains in wide-band s-wave superconductors. These results show that the order parameter contains important information regarding quantum phase transitions and their topological nature, underscoring the importance of self-consistency in such studies.

We create two molecular devices with superconducting junctions, using nickelocene molecules, single Fe atoms, and Pb electrodes at low temperature. We find contrasting behavior based on the coordination of the Fe atom: one device shows low-bias features in its differential conductance due to the superposition of multiple Andreev reflections (MAR) and Fe-induced in-gap states. The other reveals interference between MAR and in-gap states, showcasing the diversity achievable in atomically engineered devices with identical components.

This text describes experiences gained across a two-year test period during which two generations of Generative Artificial Intelligence (A.I.) systems were incorporated into an interdisciplinary, university level course on A.I. for art and design practices. The text uses the results from the courses to reflect on new opportunities for generative systems in art and design, while considering traps and limits., Comment: 20 pages, 3 tables, 4 images

We present results from our deep Far-ultraviolet (FUV) survey using {\em AstroSat}/UVIT of a filamentary structure at $z$ $\sim$ $0.072$. A total of four filaments comprising 58 galaxies were probed in our study. We detect 18 filament galaxies in our FUV observation. All filament galaxies are further classified based on their photometric color, nuclear activity, and morphology. The filaments contain galaxies with mixed stellar population types and structures. We do not detect galaxies in our UVIT survey up to a distance of 0.4~Mpc $h^{-1}$ from the filament axis, implying a lack of recent star formation in the inner region of filaments. The FUV star formation rate (SFR) for star-forming galaxies agrees well with the SFR$_{\rm 144 MHz}$ calculated using Low-Frequency Array (LOFAR) radio-continuum observations. We witness an increase in FUV specific-SFR (sSFR) of filament galaxies with increasing distance from the filament spine (D$_{\rm fil}$). The intermediate-to-high stellar mass filament galaxies were more star-forming than cluster galaxies in a fixed stellar mass bin. The FUV morphology of some filament galaxies detected in the filament outskirts (D$_{\rm fil}$ $\gtrsim$ 0.7~Mpc $h^{-1}$) is comparable to or slightly extended than their optical counterpart. The mass assembly of galaxies examined by estimating $(FUV-r)$ color gradients show that more `red-cored' galaxies reside in the outer region of the filaments. Our results prove that the likelihood of merger interaction and gas starvation increases when approaching the filament spine. We report a definitive and in-homogeneous impact of filaments on the galaxies residing inside them., Comment: Accepted for publication in ApJ

The recent advances in Deep Neural Networks (DNNs) stem from their exceptional performance across various domains. However, their inherent large size hinders deploying these networks on resource-constrained devices like edge, mobile, and IoT platforms. Strategies have emerged, from partial cloud computation offloading (split computing) to integrating early exits within DNN layers. Our work presents an innovative unified approach merging early exits and split computing. We determine the 'splitting layer', the optimal depth in the DNN for edge device computations, and whether to infer on edge device or be offloaded to the cloud for inference considering accuracy, computational efficiency, and communication costs. Also, Image classification faces diverse environmental distortions, influenced by factors like time of day, lighting, and weather. To adapt to these distortions, we introduce I-SplitEE, an online unsupervised algorithm ideal for scenarios lacking ground truths and with sequential data. Experimental validation using Caltech-256 and Cifar-10 datasets subjected to varied distortions showcases I-SplitEE's ability to reduce costs by a minimum of 55% with marginal performance degradation of at most 5%., Comment: To appear in proceedings of IEEE International Conference on Communications 2024

Surface-enhanced infrared absorption (SEIRA) spectroscopy exploits the field enhancement near nanophotonic structures for highly sensitive characterization of (bio)molecules. The vibrational signature observed in SEIRA spectra is typically interpreted as field-enhanced molecular absorption. Here we study molecular vibrations in the near field of single antennas and show that the vibrational signature can be equally well explained by field-enhanced molecular scattering. Although the infrared scattering cross section of molecules is negligible compared to their absorption cross section, the interference between the molecular-scattered field and the incident field enhances the spectral signature caused by molecular vibrational scattering by 10 orders of magnitude, thus becoming as large as that of field-enhanced molecular absorption. We provide experimental evidence that field-enhanced molecular scattering can be measured, scales in intensity with the fourth power of the local field enhancement and fully explains the vibrational signature in SEIRA spectra in both magnitude and line shape. Our work may open new paths for developing highly sensitive SEIRA sensors that exploit the presented scattering concept., Comment: 38 pages, 5 figures, 6 extended data figures

Electrically driven insulator-metal transitions in prototypical quantum materials such as VO2 offer a foundational platform for designing novel solid-state devices. Tuning the V: O stoichiometry offers a vast electronic phase space with non-trivial collective properties. Here, we report the discovery of discrete threshold switching voltages with constant threshold voltage difference between cycles in vanadium oxide crystals. The observed threshold fields over 10000 cycles are ~100X lower than that noted for stoichiometric VO2 and show unique discrete behaviour. We correlate the observed discrete memristor behaviour with the valence change mechanism and fluctuations in the chemical composition of spatially distributed VO2-VnO2n-1 complex oxide phases. Design of chemical heterogeneity in Mott insulators, therefore, offers an intriguing path to realizing low-energy neuromorphic devices., Comment: 21 pages, 4 figures

We derive the elastic field generated by multiple small-scaled inclusions distributed in a bounded set of $\mathbb{R}^3$. These inclusions are modeled with moderate values of the Lam\'e coefficients while they have a large relative mass density. These properties allow them to enjoy a sequences of resonant frequencies that can be computed via the eigenvalues of the volume integral operator having the Navier fundamental matrix as a kernel, i.e. the Navier volume operator. The dominant field, i.e. the Foldy-Lax field, models the multiple interactions between the inclusions with scattering coefficients that are inversely proportional to the difference between the used incident frequency and the already mentioned resonances. We show, in particular, that to reconstruct remotely the scattered field generated after $N$ interactions between the inclusions, one needs to use an incident frequency appropriately close to the proper resonance of the inclusions. We provide an explicit link between the order $N$ of interactions and the distance from the incident frequency to the resonance. Finally, if the cluster of the inclusions is densely distributed in a given bounded domain, then the expression of the induced dominant field suggests that the equivalent homogenized mass density can change sign depending if the used incident frequencies is smaller or larger than a certain threshold (which is explicitly given in terms of the resonant frequencies of the inclusions).

This letter presents an experimental characterization technique for assessing the performance of terahertz harmonic mixers across a wide frequency range. The total signal transfer loss of three mixers was measured in both up- and down-conversion configurations, and the conversion loss was determined through the solution of a linear system of equations. The proposed method uses LO signals with a frequency offset to ensure single sideband measurements, thereby eliminating the need for image-reject filters. The three-mixer method was verified by measurements of millimeter-wave mixers, which matched the traditional characterization method using a calibrated source and power meter. Given this successful millimeter-wave demonstration, we characterized three WM-86 Schottky diode x4-harmonic mixers from 2.2 to 3 THz. This technique presents a notable advantage for conducting broadband mixer characterizations, particularly in the terahertz frequency regime which lacks tunable, wide-band sources.

The advancement of Large Language Models (LLMs) has led to their widespread use across a broad spectrum of tasks including decision making. Prior studies have compared the decision making abilities of LLMs with those of humans from a psychological perspective. However, these studies have not always properly accounted for the sensitivity of LLMs' behavior to hyperparameters and variations in the prompt. In this study, we examine LLMs' performance on the Horizon decision making task studied by Binz and Schulz (2023) analyzing how LLMs respond to variations in prompts and hyperparameters. By experimenting on three OpenAI language models possessing different capabilities, we observe that the decision making abilities fluctuate based on the input prompts and temperature settings. Contrary to previous findings language models display a human-like exploration exploitation tradeoff after simple adjustments to the prompt., Comment: EMNLP 2023

Nonlocal evolutionary equations containing memory terms model a variety of non-Markovian processes. We present a Markovian embedding procedure for a class of nonlocal equations by utilising the spectral representation of the nonlinear memory kernel. This allows us to transform the nonlocal system to a local-in-time system in an abstract extended space. We demonstrate our embedding procedure and its efficacy for two different physical models, namely the (i) 1D walking droplet and (ii) the 1D single-phase Stefan problem.

Advances in large language models (LLMs) have driven an explosion of interest about their societal impacts. Much of the discourse around how they will impact social equity has been cautionary or negative, focusing on questions like "how might LLMs be biased and how would we mitigate those biases?" This is a vital discussion: the ways in which AI generally, and LLMs specifically, can entrench biases have been well-documented. But equally vital, and much less discussed, is the more opportunity-focused counterpoint: "what promising applications do LLMs enable that could promote equity?" If LLMs are to enable a more equitable world, it is not enough just to play defense against their biases and failure modes. We must also go on offense, applying them positively to equity-enhancing use cases to increase opportunities for underserved groups and reduce societal discrimination. There are many choices which determine the impact of AI, and a fundamental choice very early in the pipeline is the problems we choose to apply it to. If we focus only later in the pipeline -- making LLMs marginally more fair as they facilitate use cases which intrinsically entrench power -- we will miss an important opportunity to guide them to equitable impacts. Here, we highlight the emerging potential of LLMs to promote equity by presenting four newly possible, promising research directions, while keeping risks and cautionary points in clear view.

With advanced AI, while every industry is growing at rocket speed, the smart home industry has not reached the next generation. There is still a huge leap of innovation that needs to happen before we call a home a Smart home. A Smart home should predict residents' needs and fulfill them in a timely manner. One of the important tasks of maintaining a home is timely grocery tracking and supply maintenance. Grocery tracking models are very famous in the retail industry but they are nonexistent in the common household. Groceries detection in household refrigerators or storage closets is very complicated compared to retail shelving data. In this paper, home grocery tracking problem is resolved by combining retail shelving data and fruits dataset with real-time 360 view data points collected from home groceries storage. By integrating this vision-based object detection system along with supply chain and user food interest prediction systems, complete automation of groceries ordering can be achieved., Comment: 2023 IEEE International Conference on Web Intelligence and Intelligent Agent Technology (WI-IAT)

Semiconductor quantum dots are among the best sources of on-demand entangled photon pairs. The degree of entanglement, however, is generally limited by the fine structure splitting of exciton states. In this paper, we theoretically investigate the generation of polarisation-entangled photon pairs under two-photon excitation and cavity-assisted two-photon emission, both in the weak and strong cavity coupling regimes. We demonstrate and clarify that cavity coupling together with an excitation pulse reduces the degree of entanglement in three different ways. Firstly, in a strong coupling regime, cavity introduces the unequal ac-Stark shift of horizontally and vertically polarised exciton states, which results in the effective splitting of exciton states. Secondly, it induces the cross-coupling between the exciton states even in the weak coupling regime, causing the creation of unfavorable two-photon states. Finally, higher excited states of the cavity modes also contribute to the reduction of entanglement. Therefore, in the setting considered here, cavity coupling, which is generally required for the efficient collection of emitted photons, degrades the entanglement both in weak and strong coupling regimes., Comment: 14 pages, 9 figures

This report introduces a new family of multimodal models, Gemini, that exhibit remarkable capabilities across image, audio, video, and text understanding. The Gemini family consists of Ultra, Pro, and Nano sizes, suitable for applications ranging from complex reasoning tasks to on-device memory-constrained use-cases. Evaluation on a broad range of benchmarks shows that our most-capable Gemini Ultra model advances the state of the art in 30 of 32 of these benchmarks - notably being the first model to achieve human-expert performance on the well-studied exam benchmark MMLU, and improving the state of the art in every one of the 20 multimodal benchmarks we examined. We believe that the new capabilities of the Gemini family in cross-modal reasoning and language understanding will enable a wide variety of use cases. We discuss our approach toward post-training and deploying Gemini models responsibly to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.

Neural language models have become powerful tools for learning complex representations of entities in natural language processing tasks. However, their interpretability remains a significant challenge, particularly in domains like computational biology where trust in model predictions is crucial. In this work, we aim to enhance the interpretability of protein language models, specifically the state-of-the-art ESM model, by identifying and characterizing knowledge neurons - components that express understanding of key information. After fine-tuning the ESM model for the task of enzyme sequence classification, we compare two knowledge neuron selection methods that preserve a subset of neurons from the original model. The two methods, activation-based and integrated gradient-based selection, consistently outperform a random baseline. In particular, these methods show that there is a high density of knowledge neurons in the key vector prediction networks of self-attention modules. Given that key vectors specialize in understanding different features of input sequences, these knowledge neurons could capture knowledge of different enzyme sequence motifs. In the future, the types of knowledge captured by each neuron could be characterized.

This manuscript investigates wormhole solutions within the framework of extended symmetric teleparallel gravity, incorporating non-commutative geometry, and conformal symmetries. To achieve this, we examine the linear wormhole model with anisotropic fluid under Gaussian and Lorentzian distributions. The primary objective is to derive wormhole solutions while considering the influence of the shape function on model parameters under Gaussian and Lorentzian distributions. The resulting shape function satisfies all the necessary conditions for a traversable wormhole. Furthermore, we analyze the characteristics of the energy conditions and provide a detailed graphical discussion of the matter contents via energy conditions. Additionally, we explore the effect of anisotropy under Gaussian and Lorentzian distributions. Finally, we present our conclusions based on the obtained results., Comment: Accepted version in Communications in Theoretical Physics

Fractional quantum Hall (FQH) phases emerge due to strong electronic interactions and are characterized by anyonic quasiparticles, each distinguished by unique topological parameters, fractional charge, and statistics. In contrast, the integer quantum Hall (IQH) effects can be understood from the band topology of non-interacting electrons. We report a surprising super-universality of the critical behavior across all FQH and IQH transitions. Contrary to the anticipated state-dependent critical exponents, our findings reveal the same critical scaling exponent $\kappa = 0.41 \pm 0.02$ and localization length exponent $\gamma = 2.4 \pm 0.2$ for fractional and integer quantum Hall transitions. From these, we extract the value of the dynamical exponent $z\approx 1$. We have achieved this in ultra-high mobility trilayer graphene devices with a metallic screening layer close to the conduction channels. The observation of these global critical exponents across various quantum Hall phase transitions was masked in previous studies by significant sample-to-sample variation in the measured values of $\kappa$ in conventional semiconductor heterostructures, where long-range correlated disorder dominates. We show that the robust scaling exponents are valid in the limit of short-range disorder correlations., Comment: 42 pages

This paper addresses privacy concerns in multi-agent reinforcement learning (MARL), specifically within the context of supply chains where individual strategic data must remain confidential. Organizations within the supply chain are modeled as agents, each seeking to optimize their own objectives while interacting with others. As each organization's strategy is contingent on neighboring strategies, maintaining privacy of state and action-related information is crucial. To tackle this challenge, we propose a game-theoretic, privacy-preserving mechanism, utilizing a secure multi-party computation (MPC) framework in MARL settings. Our major contribution is the successful implementation of a secure MPC framework, SecFloat on EzPC, to solve this problem. However, simply implementing policy gradient methods such as MADDPG operations using SecFloat, while conceptually feasible, would be programmatically intractable. To overcome this hurdle, we devise a novel approach that breaks down the forward and backward pass of the neural network into elementary operations compatible with SecFloat , creating efficient and secure versions of the MADDPG algorithm. Furthermore, we present a learning mechanism that carries out floating point operations in a privacy-preserving manner, an important feature for successful learning in MARL framework. Experiments reveal that there is on average 68.19% less supply chain wastage in 2 PC compared to no data share, while also giving on average 42.27% better average cumulative revenue for each player. This work paves the way for practical, privacy-preserving MARL, promising significant improvements in secure computation within supply chain contexts and broadly.

Distributionally robust optimization is used to solve decision making problems under adversarial uncertainty where the distribution of the uncertainty is itself ambiguous. In this paper, we identify a class of these instances that is solvable in polynomial time by viewing it through the lens of submodularity. We show that the sharpest upper bound on the expectation of the maximum of affine functions of a random vector is computable in polynomial time if each random variable is discrete with finite support and upper bounds (respectively lower bounds) on the expected values of a finite set of submodular (respectively supermodular) functions of the random vector are specified. This adds to known polynomial time solvable instances of the multimarginal optimal transport problem and the generalized moment problem by bridging ideas from convexity in continuous optimization to submodularity in discrete optimization. In turn, we show that a class of distributionally robust optimization problems with discrete random variables is solvable in polynomial time using the ellipsoid method. When the submodular (respectively supermodular) functions are structured, the sharp bound is computable by solving a compact linear program. We illustrate this in two cases. The first is a multimarginal optimal transport problem with given univariate marginal distributions and bivariate marginals satisfying specific positive dependence orders along with an extension to incorporate higher order marginal information. The second is a discrete moment problem where a set of marginal moments of the random variables are given along with lower bounds on the cross moments of pairs of random variables. Numerical experiments show that the bounds improve by 2 to 8 percent over bounds that use only univariate information in the first case, and by 8 to 15 percent over bounds that use the first moment in the second case., Comment: 36 Pages, 6 Figures