Search

Recommender systems are widely used to suggest engaging content, and Large Language Models (LLMs) have given rise to generative recommenders. Such systems can directly generate items, including for open-set tasks like question suggestion. While the world knowledge of LLMs enable good recommendations, improving the generated content through user feedback is challenging as continuously fine-tuning LLMs is prohibitively expensive. We present a training-free approach for optimizing generative recommenders by connecting user feedback loops to LLM-based optimizers. We propose a generative explore-exploit method that can not only exploit generated items with known high engagement, but also actively explore and discover hidden population preferences to improve recommendation quality. We evaluate our approach on question generation in two domains (e-commerce and general knowledge), and model user feedback with Click Through Rate (CTR). Experiments show our LLM-based explore-exploit approach can iteratively improve recommendations, and consistently increase CTR. Ablation analysis shows that generative exploration is key to learning user preferences, avoiding the pitfalls of greedy exploit-only approaches. A human evaluation strongly supports our quantitative findings., Comment: Accepted at ACL 2024 Main Proceedings

Conversational Task Assistants (CTAs) guide users in performing a multitude of activities, such as making recipes. However, ensuring that interactions remain engaging, interesting, and enjoyable for CTA users is not trivial, especially for time-consuming or challenging tasks. Grounded in psychological theories of human interest, we propose to engage users with contextual and interesting statements or facts during interactions with a multi-modal CTA, to reduce fatigue and task abandonment before a task is complete. To operationalize this idea, we train a high-performing classifier (82% F1-score) to automatically identify relevant and interesting facts for users. We use it to create an annotated dataset of task-specific interesting facts for the domain of cooking. Finally, we design and validate a dialogue policy to incorporate the identified relevant and interesting facts into a conversation, to improve user engagement and task completion. Live testing on a leading multi-modal voice assistant shows that 66% of the presented facts were received positively, leading to a 40% gain in the user satisfaction rating, and a 37% increase in conversation length. These findings emphasize that strategically incorporating interesting facts into the CTA experience can promote real-world user participation for guided task interactions., Comment: 10 pages, 1 figure

Large Language Models (LLMs) operating in 0-shot or few-shot settings achieve competitive results in Text Classification tasks. In-Context Learning (ICL) typically achieves better accuracy than the 0-shot setting, but it pays in terms of efficiency, due to the longer input prompt. In this paper, we propose a strategy to make LLMs as efficient as 0-shot text classifiers, while getting comparable or better accuracy than ICL. Our solution targets the low resource setting, i.e., when only 4 examples per class are available. Using a single LLM and few-shot real data we perform a sequence of generation, filtering and Parameter-Efficient Fine-Tuning steps to create a robust and efficient classifier. Experimental results show that our approach leads to competitive results on multiple text classification datasets., Comment: Accepted at LREC-COLING 2024

Many Natural Language Generation (NLG) tasks aim to generate a single output text given an input prompt. Other settings require the generation of multiple texts, e.g., for Synthetic Traffic Generation (STG). This generation task is crucial for training and evaluating QA systems as well as conversational agents, where the goal is to generate multiple questions or utterances resembling the linguistic variability of real users. In this paper, we show that common NLG metrics, like BLEU, are not suitable for evaluating STG. We propose and evaluate several metrics designed to compare the generated traffic to the distribution of real user texts. We validate our metrics with an automatic procedure to verify whether they capture different types of quality issues of generated data; we also run human annotations to verify the correlation with human judgements. Experiments on three tasks, i.e., Shopping Utterance Generation, Product Question Generation and Query Auto Completion, demonstrate that our metrics are effective for evaluating STG tasks, and improve the agreement with human judgement up to 20% with respect to common NLG metrics. We believe these findings can pave the way towards better solutions for estimating the representativeness of synthetic text data.

A recent Maxey--Riley theory for \sarg raft motion, which models a raft as a network of elastically interacting finite-size, buoyant particles, predicts the carrying flow velocity to be given by the weighted sum of the water and air velocities $(1-\alpha)\mathbf{v} + \alpha \mathbf w$. The theory provides a closed formula for parameter $\alpha$, referred to as \emph{windage}, depending on water-to-particle-density ratio or buoyancy ($\delta$). From a series of laboratory experiments in an air--water stream flume facility under controlled conditions, we estimate $\alpha$ ranging from 0.02 to 0.96\pct. On average, our windage estimates can be up to 9 times smaller than considered in conventional \emph{Sargassum} raft transport modeling, wherein it is customary to add a fraction of $\mathbf w$ to $\mathbf{v}$ chosen in an ad-hoc piecemeal manner. Using the formula provided by the Maxey--Riley theory, we estimate $\delta$ ranging from 1.00 to 1.49. This is consistent with direct $\delta$ measurements, ranging from 0.9 to 1.25, which provide support for our $\alpha$ estimation.

The adoption of voice assistants like Alexa or Siri has grown rapidly, allowing users to instantly access information via voice search. Query suggestion is a standard feature of screen-based search experiences, allowing users to explore additional topics. However, this is not trivial to implement in voice-based settings. To enable this, we tackle the novel task of suggesting questions with compact and natural voice hints to allow users to ask follow-up questions. We define the task, ground it in syntactic theory and outline linguistic desiderata for spoken hints. We propose baselines and an approach using sequence-to-sequence Transformers to generate spoken hints from a list of questions. Using a new dataset of 6681 input questions and human written hints, we evaluated the models with automatic metrics and human evaluation. Results show that a naive approach of concatenating suggested questions creates poor voice hints. Our approach, which applies a linguistically-motivated pretraining task was strongly preferred by humans for producing the most natural hints., Comment: Accepted as Long Paper at EMNLP'23 Findings

Multi-Task Learning (MTL) is widely-accepted in Natural Language Processing as a standard technique for learning multiple related tasks in one model. Training an MTL model requires having the training data for all tasks available at the same time. As systems usually evolve over time, (e.g., to support new functionalities), adding a new task to an existing MTL model usually requires retraining the model from scratch on all the tasks and this can be time-consuming and computationally expensive. Moreover, in some scenarios, the data used to train the original training may be no longer available, for example, due to storage or privacy concerns. In this paper, we approach the problem of incrementally expanding MTL models' capability to solve new tasks over time by distilling the knowledge of an already trained model on n tasks into a new one for solving n+1 tasks. To avoid catastrophic forgetting, we propose to exploit unlabeled data from the same distributions of the old tasks. Our experiments on publicly available benchmarks show that such a technique dramatically benefits the distillation by preserving the already acquired knowledge (i.e., preventing up to 20% performance drops on old tasks) while obtaining good performance on the incrementally added tasks. Further, we also show that our approach is beneficial in practical settings by using data from a leading voice assistant., Comment: KDD 2022

Since its inception in 2016, the Alexa Prize program has enabled hundreds of university students to explore and compete to develop conversational agents through the SocialBot Grand Challenge. The goal of the challenge is to build agents capable of conversing coherently and engagingly with humans on popular topics for 20 minutes, while achieving an average rating of at least 4.0/5.0. However, as conversational agents attempt to assist users with increasingly complex tasks, new conversational AI techniques and evaluation platforms are needed. The Alexa Prize TaskBot challenge, established in 2021, builds on the success of the SocialBot challenge by introducing the requirements of interactively assisting humans with real-world Cooking and Do-It-Yourself tasks, while making use of both voice and visual modalities. This challenge requires the TaskBots to identify and understand the user's need, identify and integrate task and domain knowledge into the interaction, and develop new ways of engaging the user without distracting them from the task at hand, among other challenges. This paper provides an overview of the TaskBot challenge, describes the infrastructure support provided to the teams with the CoBot Toolkit, and summarizes the approaches the participating teams took to overcome the research challenges. Finally, it analyzes the performance of the competing TaskBots during the first year of the competition., Comment: 14 pages, Proceedings of Alexa Prize Taskbot (Alexa Prize 2021)

We propose and demonstrate an experimental implementation for the observation of magnetic fields from spatial features of absorption profiles in a warm atomic vapor. A radially polarized vector beam that traverses an atomic vapor will generate an absorption pattern with petal-like structure by the mediation of a transverse magnetic field (TMF). The spatial absorption pattern rotates when the azimuthal angle of the TMF is changed, while its contrast decreases when the longitudinal component of the magnetic field increases. By analyzing the intensity distribution of the transmitted pattern we can determine the magnetic field strength. Our work provides a framework for investigating three-dimensional magnetic field distributions based on atoms.

We describe and demonstrate how 3D magnetic field alignment can be inferred from single absorption images of an atomic cloud. While optically pumped magnetometers conventionally rely on temporal measurement of the Larmor precession of atomic dipoles, here a cold atomic vapour provides a spatial interface between vector light and external magnetic fields. Using a vector vortex beam, we inscribe structured atomic spin polarisation in a cloud of cold rubidium atoms, and record images of the resulting absorption patterns. The polar angle of an external magnetic field can be deduced with spatial Fourier analysis. This effect presents an alternative concept for detecting magnetic vector fields, and demonstrates, more generally, how introducing spatial phases between atomic energy levels can translate transient effects to the spatial domain.

Research on spatially-structured light has seen an explosion in activity over the past decades, powered by technological advances for generating such light, and driven by questions of fundamental science as well as engineering applications. In this review we highlight work on the interaction of vector light fields with atoms, and matter in general. This vibrant research area explores the full potential of light, with clear benefits for classical as well as quantum applications.

The widespread use of conversational and question answering systems made it necessary to improve the performances of speaker intent detection and understanding of related semantic slots, i.e., Spoken Language Understanding (SLU). Often, these tasks are approached with supervised learning methods, which needs considerable labeled datasets. This paper presents the first Italian dataset for SLU. It is derived through a semi-automatic procedure and is used as a benchmark of various open source and commercial systems.

Research on container loading problems has been proved effective in increasing the filling rate of containers in different practical situations. However, the broader logistic context might pressure the loading process, leading to sub-optimal solutions. Some facilities like cross-docks have reduced storage space which might force early loading activities. We propose a container loading problem which accounts for this limited storage by explicitly considering the schedule of arrival for the boxes and the departure time of the trucks. Also, we design a framework which handles the geometric and temporal characteristics of the problem separately, enabling the use of methods found in the literature for solving the extended problem. Our framework can handle uncertainty in the schedule and be used to quantify the impact of delays on capacity utilization and departure time of trucks.

Intent Detection and Slot Filling are two pillar tasks in Spoken Natural Language Understanding. Common approaches adopt joint Deep Learning architectures in attention-based recurrent frameworks. In this work, we aim at exploiting the success of "recurrence-less" models for these tasks. We introduce Bert-Joint, i.e., a multi-lingual joint text classification and sequence labeling framework. The experimental evaluation over two well-known English benchmarks demonstrates the strong performances that can be obtained with this model, even when few annotated data is available. Moreover, we annotated a new dataset for the Italian language, and we observed similar performances without the need for changing the model.

Vector vortex beams possess a topological property that derives both from the spatially varying amplitude of the field and also from its varying polarization. This property arises as a consequence of the inherent Skyrmionic nature of such beams and is quantified by the associated Skyrmion number, which embodies a topological property of the beam. We illustrate this idea for some of the simplest vector beams and discuss the physical significance of the Skyrmion number in this context., Comment: 6 pages, 6 figures

Understanding and probing phase transitions in non-equilibrium systems is an ongoing challenge in physics. A particular instance are phase transitions that occur between a non-fluctuating absorbing phase, e.g., an extinct population, and one in which the relevant order parameter, such as the population density, assumes a finite value. Here we report the observation of signatures of such a non-equilibrium phase transition in an open driven quantum system. In our experiment rubidium atoms in a quasi one-dimensional cold disordered gas are laser-excited to Rydberg states under so-called facilitation conditions. This conditional excitation process competes with spontaneous decay and leads to a crossover between a stationary state with no excitations and one with a finite number of excitations. We relate the underlying physics to that of an absorbing state phase transition in the presence of a field (i.e. off-resonant excitation processes) which slightly offsets the system from criticality. We observe a characteristic power-law scaling of the Rydberg excitation density as well as increased fluctuations close to the transition point. Furthermore, we argue that the observed transition relies on the presence of atomic motion which introduces annealed disorder into the system and enables the formation of long-ranged correlations. Our study paves the road for future investigations into the largely unexplored physics of non-equilibrium phase transitions in open many-body quantum systems., Comment: 14 pages, 8 figures

An interesting empirical result in the assembly line literature states that slightly unbalanced assembly lines (in the format of a bowl - with central stations less loaded than the external ones) present higher throughputs than perfectly balanced ones. This effect is known as the bowl phenomenon. In this study, we analyze the presence of this phenomenon in assembly lines with heterogeneous workers (most of them, workers with disabilities). For this purpose, we modify an existing model for the assembly line worker assignment and balancing problem in order to generate configurations exhibiting the desired format. These configurations are implemented in a stochastic simulation model and the obtained results are analyzed. The findings obtained here confirm the existence of the bowl phenomenon in such situations and contributes to deepen our knowledge in this empirical phenomenon.

This paper describes the development of the Detector Array Assembly (DAA) structure for the Advanced Topographic Laser Altimeter System (ATLAS) on the ICESat2 Mission built at NASA Goddard Space Flight Center (GSFC). It describes the challenges faced in designing a structure to protect sensitive detectors with limited data available for the detectors and the environment, including shock. The paper shows the successful re-design and testing that was performed to allow for a design to handle launch, shock and thermal environments with flexibility to incorporate design changes.

Recent advancements in composite materials technologies have broken further from traditional designs and require advanced instrumentation and analysis capabilities. Success or failure is highly dependent on design analysis and manufacturing processes. By monitoring smart structures throughout manufacturing and service life, residual and operational stresses can be assessed and structural integrity maintained. Composite smart structures can be manufactured by integrating fiber optic sensors into existing composite materials processes such as ply layup, filament winding and three-dimensional weaving. In this work optical fiber was integrated into 3D woven composite parts at a commercial woven products manufacturing facility. The fiber was then used to monitor the structures during a VARTM manufacturing process, and subsequent static and dynamic testing. Low cost telecommunications-grade optical fiber acts as the sensor using a high resolution commercial Optical Frequency Domain Reflectometer (OFDR) system providing distributed strain measurement at spatial resolutions as low as 2mm. Strain measurements using the optical fiber sensors are correlated to resistive strain gage measurements during static structural loading. Keywords: fiber optic, distributed strain sensing, Rayleigh scatter, optical frequency domain reflectometry

The Laser Interferometer Space Antenna (LISA) mission. a space based gravitational wave detector. uses laser metrology to measure distance fluctuations between proof masses aboard three spacecraft. LISA is unique from a mission design perspective in that the three spacecraft and their associated operations form one distributed science instrument. unlike more conventional missions where an instrument is a component of an individual spacecraft. The design of the LISA spacecraft is also tightly coupled to the design and requirements of the scientific payload; for this reason it is often referred to as a "sciencecraft." Here we describe some of the unique features of the LISA spacecraft design that help create the quiet environment necessary for gravitational wave observations.

A report describes experiments directed toward the development of paraffin-based hybrid rocket fuels that burn at regression rates greater than those of conventional hybrid rocket fuels like hydroxyl-terminated butadiene. The basic approach followed in this development is to use materials such that a hydrodynamically unstable liquid layer forms on the melting surface of a burning fuel body. Entrainment of droplets from the liquid/gas interface can substantially increase the rate of fuel mass transfer, leading to surface regression faster than can be achieved using conventional fuels. The higher regression rate eliminates the need for the complex multi-port grain structures of conventional solid rocket fuels, making it possible to obtain acceptable performance from single-port structures. The high-regression-rate fuels contain no toxic or otherwise hazardous components and can be shipped commercially as non-hazardous commodities. Among the experiments performed on these fuels were scale-up tests using gaseous oxygen. The data from these tests were found to agree with data from small-scale, low-pressure and low-mass-flux laboratory tests and to confirm the expectation that these fuels would burn at high regression rates, chamber pressures, and mass fluxes representative of full-scale rocket motors.