Search

Hypervelocity white dwarfs (HVWDs) are stellar remnants moving at speeds exceeding the Milky Way's escape velocity. The origins of the fastest HVWDs are enigmatic, with proposed formation scenarios facing challenges explaining both their extreme velocities and observed properties. Here we report a three-dimensional hydrodynamic simulation of a merger between two hybrid helium-carbon-oxygen white dwarfs (HeCO WDs with masses of 0.68 and 0.62 M$_\odot$). We find that the merger leads to a partial disruption of the secondary WD, coupled with a double-detonation explosion of the primary WD. This launches the remnant core of the secondary WD at a speed of $~2000$ km s$^{-1}$, consistent with observed HVWDs. The low mass of the ejected remnant and its heating from the primary WD's ejecta explain the observed luminosities and temperatures of hot HVWDs, which are otherwise difficult to reconcile with previous models. This discovery establishes a new formation channel for HVWDs and points to a previously unrecognized pathway for producing peculiar Type Ia supernovae and faint explosive transients., Comment: 13 pages, 6 figures. Comments welcome!

Soundness of a type system is a fundemental property that guarantees that no operation that is not supported by a value will be performed on that value at run time. A type checker for a sound type system is expected to issue a warning on every type error. While soundness is a desirable property for many practical applications, in 2016, Amin and Tate presented the first unsoundness proof for two major industry languages: Java and Scala. This proof relied on use-site variance and implicit null values. We present an unsoundness proof for Kotlin, another emerging industry language, which relies on a previously unknown unsound combination of language features. Kotlin does not have implicit null values, meaning that the proof by Amin and Tate would not work for Kotlin. Our new proof, which is an infringing code snippet, utilizes Kotlin's \emph{declaration-site} variance specification and does not require implicit null values. We present this counterexample to soundness in full along with detailed explanations of every step. Finally, we present a thorough discussion on precisely which language features cause this issue, as well as how Kotlin's compiler can be patched to fix it.

Despite their wide adoption, the biases and unintended behaviors of language models remain poorly understood. In this paper, we identify and characterize a phenomenon never discussed before, which we call semantic leakage, where models leak irrelevant information from the prompt into the generation in unexpected ways. We propose an evaluation setting to detect semantic leakage both by humans and automatically, curate a diverse test suite for diagnosing this behavior, and measure significant semantic leakage in 13 flagship models. We also show that models exhibit semantic leakage in languages besides English and across different settings and generation scenarios. This discovery highlights yet another type of bias in language models that affects their generation patterns and behavior.

The quantum approximate optimisation algorithm (QAOA) is at the core of many scenarios that aim to combine the power of quantum computers and classical high-performance computing appliances for combinatorial optimisation. Several obstacles challenge concrete benefits now and in the foreseeable future: Imperfections quickly degrade algorithmic performance below practical utility; overheads arising from alternating between classical and quantum primitives can counter any advantage; and the choice of parameters or algorithmic variant can substantially influence runtime and result quality. Selecting the optimal combination is a non-trivial issue, as it not only depends on user requirements, but also on details of the hardware and software stack. Appropriate automation can lift the burden of choosing optimal combinations for end-users: They should not be required to understand technicalities like differences between QAOA variants, required number of QAOA layers, or necessary measurement samples. Yet, they should receive best-possible satisfaction of their non-functional requirements, be it performance or other. We determine factors that affect solution quality and temporal behaviour of four QAOA variants using comprehensive density-matrix-based simulations targeting three widely studied optimisation problems. Our simulations consider ideal quantum computation, and a continuum of scenarios troubled by realistic imperfections. Our quantitative results, accompanied by a comprehensive reproduction package, show strong differences between QAOA variants that can be pinpointed to narrow and specific effects. We identify influential co-variables and relevant non-functional quality goals that, we argue, mark the relevant ingredients for designing appropriate software engineering abstraction mechanisms and automated tool-chains for devising quantum solutions from high-level problem specifications., Comment: To be published as a workshop paper at QCE 2024 (WIHPQC 2024)

In multilingual settings, non-Latin scripts and low-resource languages are usually disadvantaged in terms of language models' utility, efficiency, and cost. Specifically, previous studies have reported multiple modeling biases that the current tokenization algorithms introduce to non-Latin script languages, the main one being over-segmentation. In this work, we propose MAGNET; multilingual adaptive gradient-based tokenization to reduce over-segmentation via adaptive gradient-based subword tokenization. MAGNET learns to predict segment boundaries between byte tokens in a sequence via sub-modules within the model, which act as internal boundary predictors (tokenizers). Previous gradient-based tokenization methods aimed for uniform compression across sequences by integrating a single boundary predictor during training and optimizing it end-to-end through stochastic reparameterization alongside the next token prediction objective. However, this approach still results in over-segmentation for non-Latin script languages in multilingual settings. In contrast, MAGNET offers a customizable architecture where byte-level sequences are routed through language-script-specific predictors, each optimized for its respective language script. This modularity enforces equitable segmentation granularity across different language scripts compared to previous methods. Through extensive experiments, we demonstrate that in addition to reducing segmentation disparities, MAGNET also enables faster language modelling and improves downstream utility.

Customizing text-to-image (T2I) models has seen tremendous progress recently, particularly in areas such as personalization, stylization, and conditional generation. However, expanding this progress to video generation is still in its infancy, primarily due to the lack of customized video data. In this work, we introduce Still-Moving, a novel generic framework for customizing a text-to-video (T2V) model, without requiring any customized video data. The framework applies to the prominent T2V design where the video model is built over a text-to-image (T2I) model (e.g., via inflation). We assume access to a customized version of the T2I model, trained only on still image data (e.g., using DreamBooth or StyleDrop). Naively plugging in the weights of the customized T2I model into the T2V model often leads to significant artifacts or insufficient adherence to the customization data. To overcome this issue, we train lightweight $\textit{Spatial Adapters}$ that adjust the features produced by the injected T2I layers. Importantly, our adapters are trained on $\textit{"frozen videos"}$ (i.e., repeated images), constructed from image samples generated by the customized T2I model. This training is facilitated by a novel $\textit{Motion Adapter}$ module, which allows us to train on such static videos while preserving the motion prior of the video model. At test time, we remove the Motion Adapter modules and leave in only the trained Spatial Adapters. This restores the motion prior of the T2V model while adhering to the spatial prior of the customized T2I model. We demonstrate the effectiveness of our approach on diverse tasks including personalized, stylized, and conditional generation. In all evaluated scenarios, our method seamlessly integrates the spatial prior of the customized T2I model with a motion prior supplied by the T2V model., Comment: Webpage: https://still-moving.github.io/ | Video: https://www.youtube.com/watch?v=U7UuV_VIjnA

Yor\`ub\'a an African language with roughly 47 million speakers encompasses a continuum with several dialects. Recent efforts to develop NLP technologies for African languages have focused on their standard dialects, resulting in disparities for dialects and varieties for which there are little to no resources or tools. We take steps towards bridging this gap by introducing a new high-quality parallel text and speech corpus YOR\`ULECT across three domains and four regional Yor\`ub\'a dialects. To develop this corpus, we engaged native speakers, travelling to communities where these dialects are spoken, to collect text and speech data. Using our newly created corpus, we conducted extensive experiments on (text) machine translation, automatic speech recognition, and speech-to-text translation. Our results reveal substantial performance disparities between standard Yor\`ub\'a and the other dialects across all tasks. However, we also show that with dialect-adaptive finetuning, we are able to narrow this gap. We believe our dataset and experimental analysis will contribute greatly to developing NLP tools for Yor\`ub\'a and its dialects, and potentially for other African languages, by improving our understanding of existing challenges and offering a high-quality dataset for further development. We release YOR\`ULECT dataset and models publicly under an open license.

Reinforcement Learning from Human Feedback (RLHF) has become the standard approach for aligning Large Language Models (LLMs) with human preferences, allowing LLMs to demonstrate remarkable abilities in various tasks. Existing methods work by emulating the preferences at the single decision (turn) level, limiting their capabilities in settings that require planning or multi-turn interactions to achieve a long-term goal. In this paper, we address this issue by developing novel methods for Reinforcement Learning (RL) from preference feedback between two full multi-turn conversations. In the tabular setting, we present a novel mirror-descent-based policy optimization algorithm for the general multi-turn preference-based RL problem, and prove its convergence to Nash equilibrium. To evaluate performance, we create a new environment, Education Dialogue, where a teacher agent guides a student in learning a random topic, and show that a deep RL variant of our algorithm outperforms RLHF baselines. Finally, we show that in an environment with explicit rewards, our algorithm recovers the same performance as a reward-based RL baseline, despite relying solely on a weaker preference signal.

Portable heart rate monitoring (HRM) systems based on electrocardiograms (ECGs) have become increasingly crucial for preventing lifestyle diseases. For such portable systems, minimizing power consumption and sampling rate is critical due to the substantial data generated during long-term ECG monitoring. The variable pulse-width finite rate of innovation (VPW-FRI) framework provides an effective solution for low-rate sampling and compression of ECG signals. We develop a time-based sub-Nyquist sampling and reconstruction method for ECG signals specifically designed for HRM applications. Our approach harnesses the integrate-and-fire time-encoding machine (IF-TEM) as a power-efficient, time-based, asynchronous sampler, generating a sequence of time instants without the need for a global clock. The ECG signal is represented as a linear combination of VPW-FRI pulses, which is then subjected to pre-filtering before being sampled by the IF-TEM sampler. A compactly supported robust filter with a frequency-domain alias cancellation condition is used to combat the effects of noise. Our reconstruction process involves consecutive partial summations of discrete representations of the input signal derived from the series of time encodings, further enhancing the accuracy of the reconstructed ECG signals. Additionally, we introduce an IF-TEM sampling hardware system for ECG signals, implemented using an analog filter device. The firing rate is 42-80Hz, equivalent to approximately 0.025-0.05 of the Nyquist rate. Our hardware validation bridges the gap between theory and practice and demonstrates the robust performance and practical applicability of our approach in accurately monitoring heart rates and reconstructing ECG signals.

This investigation examined the relationships among scene complexity, workload, presence, and cybersickness in virtual reality (VR) environments. Numerous factors can influence the overall VR experience, and existing research on this matter is not yet conclusive, warranting further investigation. In this between-subjects experimental setup, 44 participants engaged in the Pendulum Chair game, with half exposed to a simple scene with lower optic flow and lower familiarity, and the remaining half to a complex scene characterized by higher optic flow and greater familiarity. The study measured the dependent variables workload, presence, and cybersickness and analyzed their correlations. Equivalence testing was also used to compare the simple and complex environments. Results revealed that despite the visible differences between the environments, within the 10% boundaries of the maximum possible value for workload and presence, and 13.6% of the maximum SSQ value, a statistically significant equivalence was observed between the simple and complex scenes. Additionally, a moderate, negative correlation emerged between workload and SSQ scores. The findings suggest two key points: (1) the nature of the task can mitigate the impact of scene complexity factors such as optic flow and familiarity, and (2) the correlation between workload and cybersickness may vary, showing either a positive or negative relationship.

A major consideration in multilingual language modeling is how to best represent languages with diverse vocabularies and scripts. Although contemporary text encoding methods cover most of the world's writing systems, they exhibit bias towards the high-resource languages of the Global West. As a result, texts of underrepresented languages tend to be segmented into long sequences of linguistically meaningless units. To address the disparities, we introduce a new paradigm that encodes the same information with segments of consistent size across diverse languages. Our encoding convention (MYTE) is based on morphemes, as their inventories are more balanced across languages than characters, which are used in previous methods. We show that MYTE produces shorter encodings for all 99 analyzed languages, with the most notable improvements for non-European languages and non-Latin scripts. This, in turn, improves multilingual LM performance and diminishes the perplexity gap throughout diverse languages.

We carry out three-dimensional computations of the accretion rate onto an object (of size $R_{\rm sink}$ and mass $m$) as it moves through a uniform medium at a subsonic speed $v_{\infty}$. The object is treated as a fully-absorbing boundary (e.g. a black hole). In contrast to early conjectures, we show that when $R_{\rm sink}\ll R_{A}=2Gm/v^2$ the accretion rate is independent of $v_{\infty}$ and only depends on the entropy of the ambient medium, its adiabatic index, and $m$. Our numerical simulations are conducted using two different numerical schemes via the Athena++ and Arepo hydrodynamics solvers, which reach nearly identical steady-state solutions. We find that pressure gradients generated by the isentropic compression of the flow near the accretor are sufficient to suspend much of the surrounding gas in a near-hydrostatic equilibrium, just as predicted from the spherical Bondi-Hoyle calculation. Indeed, the accretion rates for steady flow match the Bondi-Hoyle rate, and are indicative of isentropic flow for subsonic motion where no shocks occur. We also find that the accretion drag may be predicted using the Safronov number, $\Theta=R_{A}/R_{\rm sink}$, and is much less than the dynamical friction for sufficiently small accretors ($R_{\rm sink}\ll R_{A}$)., Comment: 14 pages, 7 figures. Submitted to ApJ. Comments welcome!

Editing signals using large pre-trained models, in a zero-shot manner, has recently seen rapid advancements in the image domain. However, this wave has yet to reach the audio domain. In this paper, we explore two zero-shot editing techniques for audio signals, which use DDPM inversion with pre-trained diffusion models. The first, which we coin ZEro-shot Text-based Audio (ZETA) editing, is adopted from the image domain. The second, named ZEro-shot UnSupervized (ZEUS) editing, is a novel approach for discovering semantically meaningful editing directions without supervision. When applied to music signals, this method exposes a range of musically interesting modifications, from controlling the participation of specific instruments to improvisations on the melody. Samples and code can be found in https://hilamanor.github.io/AudioEditing/ ., Comment: Accepted for ICML 2024; Examples and code available in https://hilamanor.github.io/AudioEditing/

We introduce Lumiere -- a text-to-video diffusion model designed for synthesizing videos that portray realistic, diverse and coherent motion -- a pivotal challenge in video synthesis. To this end, we introduce a Space-Time U-Net architecture that generates the entire temporal duration of the video at once, through a single pass in the model. This is in contrast to existing video models which synthesize distant keyframes followed by temporal super-resolution -- an approach that inherently makes global temporal consistency difficult to achieve. By deploying both spatial and (importantly) temporal down- and up-sampling and leveraging a pre-trained text-to-image diffusion model, our model learns to directly generate a full-frame-rate, low-resolution video by processing it in multiple space-time scales. We demonstrate state-of-the-art text-to-video generation results, and show that our design easily facilitates a wide range of content creation tasks and video editing applications, including image-to-video, video inpainting, and stylized generation., Comment: Webpage: https://lumiere-video.github.io/ | Video: https://www.youtube.com/watch?v=wxLr02Dz2Sc

Despite their popularity in non-English NLP, multilingual language models often underperform monolingual ones due to inter-language competition for model parameters. We propose Cross-lingual Expert Language Models (X-ELM), which mitigate this competition by independently training language models on subsets of the multilingual corpus. This process specializes X-ELMs to different languages while remaining effective as a multilingual ensemble. Our experiments show that when given the same compute budget, X-ELM outperforms jointly trained multilingual models across all considered languages and that these gains transfer to downstream tasks. X-ELM provides additional benefits over performance improvements: new experts can be iteratively added, adapting X-ELM to new languages without catastrophic forgetting. Furthermore, training is asynchronous, reducing the hardware requirements for multilingual training and democratizing multilingual modeling., Comment: EMNLP 2024

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.

We introduce Universal NER (UNER), an open, community-driven project to develop gold-standard NER benchmarks in many languages. The overarching goal of UNER is to provide high-quality, cross-lingually consistent annotations to facilitate and standardize multilingual NER research. UNER v1 contains 18 datasets annotated with named entities in a cross-lingual consistent schema across 12 diverse languages. In this paper, we detail the dataset creation and composition of UNER; we also provide initial modeling baselines on both in-language and cross-lingual learning settings. We release the data, code, and fitted models to the public., Comment: NAACL 2024 Camera-ready

Image restoration problems are typically ill-posed in the sense that each degraded image can be restored in infinitely many valid ways. To accommodate this, many works generate a diverse set of outputs by attempting to randomly sample from the posterior distribution of natural images given the degraded input. Here we argue that this strategy is commonly of limited practical value because of the heavy tail of the posterior distribution. Consider for example inpainting a missing region of the sky in an image. Since there is a high probability that the missing region contains no object but clouds, any set of samples from the posterior would be entirely dominated by (practically identical) completions of sky. However, arguably, presenting users with only one clear sky completion, along with several alternative solutions such as airships, birds, and balloons, would better outline the set of possibilities. In this paper, we initiate the study of meaningfully diverse image restoration. We explore several post-processing approaches that can be combined with any diverse image restoration method to yield semantically meaningful diversity. Moreover, we propose a practical approach for allowing diffusion based image restoration methods to generate meaningfully diverse outputs, while incurring only negligent computational overhead. We conduct extensive user studies to analyze the proposed techniques, and find the strategy of reducing similarity between outputs to be significantly favorable over posterior sampling. Code and examples are available at https://noa-cohen.github.io/MeaningfulDiversityInIR., Comment: Accepted for ICLR 2024. Code and examples are available at https://noa-cohen.github.io/MeaningfulDiversityInIR

The translation of ambiguous text presents a challenge for translation systems, as it requires using the surrounding context to disambiguate the intended meaning as much as possible. While prior work has studied ambiguities that result from different grammatical features of the source and target language, we study semantic ambiguities that exist in the source (English in this work) itself. In particular, we focus on idioms that are open to both literal and figurative interpretations (e.g., goose egg), and collect TIDE, a dataset of 512 pairs of English sentences containing idioms with disambiguating context such that one is literal (it laid a goose egg) and another is figurative (they scored a goose egg, as in a score of zero). In experiments, we compare MT-specific models and language models for (i) their preference when given an ambiguous subsentence, (ii) their sensitivity to disambiguating context, and (iii) the performance disparity between figurative and literal source sentences. We find that current MT models consistently translate English idioms literally, even when the context suggests a figurative interpretation. On the other hand, LMs are far more context-aware, although there remain disparities across target languages. Our findings underline the potential of LMs as a strong backbone for context-aware translation., Comment: EMNLP 2023 Findings

The task of open-vocabulary object-centric image retrieval involves the retrieval of images containing a specified object of interest, delineated by an open-set text query. As working on large image datasets becomes standard, solving this task efficiently has gained significant practical importance. Applications include targeted performance analysis of retrieved images using ad-hoc queries and hard example mining during training. Recent advancements in contrastive-based open vocabulary systems have yielded remarkable breakthroughs, facilitating large-scale open vocabulary image retrieval. However, these approaches use a single global embedding per image, thereby constraining the system's ability to retrieve images containing relatively small object instances. Alternatively, incorporating local embeddings from detection pipelines faces scalability challenges, making it unsuitable for retrieval from large databases. In this work, we present a simple yet effective approach to object-centric open-vocabulary image retrieval. Our approach aggregates dense embeddings extracted from CLIP into a compact representation, essentially combining the scalability of image retrieval pipelines with the object identification capabilities of dense detection methods. We show the effectiveness of our scheme to the task by achieving significantly better results than global feature approaches on three datasets, increasing accuracy by up to 15 mAP points. We further integrate our scheme into a large scale retrieval framework and demonstrate our method's advantages in terms of scalability and interpretability., Comment: BMVC 2023