Search

We propose a new programming language called ALTA and a compiler that can map ALTA programs to Transformer weights. ALTA is inspired by RASP, a language proposed by Weiss et al. (2021), and Tracr (Lindner et al., 2023), a compiler from RASP programs to Transformer weights. ALTA complements and extends this prior work, offering the ability to express loops and to compile programs to Universal Transformers, among other advantages. ALTA allows us to constructively show how Transformers can represent length-invariant algorithms for computing parity and addition, as well as a solution to the SCAN benchmark of compositional generalization tasks, without requiring intermediate scratchpad decoding steps. We also propose tools to analyze cases where the expressibility of an algorithm is established, but end-to-end training on a given training set fails to induce behavior consistent with the desired algorithm. To this end, we explore training from ALTA execution traces as a more fine-grained supervision signal. This enables additional experiments and theoretical analyses relating the learnability of various algorithms to data availability and modeling decisions, such as positional encodings. We make the ALTA framework -- language specification, symbolic interpreter, and weight compiler -- available to the community to enable further applications and insights.

Efficiently retrieving and synthesizing information from large-scale multimodal collections has become a critical challenge. However, existing video retrieval datasets suffer from scope limitations, primarily focusing on matching descriptive but vague queries with small collections of professionally edited, English-centric videos. To address this gap, we introduce $\textbf{MultiVENT 2.0}$, a large-scale, multilingual event-centric video retrieval benchmark featuring a collection of more than 218,000 news videos and 3,906 queries targeting specific world events. These queries specifically target information found in the visual content, audio, embedded text, and text metadata of the videos, requiring systems leverage all these sources to succeed at the task. Preliminary results show that state-of-the-art vision-language models struggle significantly with this task, and while alternative approaches show promise, they are still insufficient to adequately address this problem. These findings underscore the need for more robust multimodal retrieval systems, as effective video retrieval is a crucial step towards multimodal content understanding and generation tasks.

Quantum error correction (QEC) will be essential to achieve the accuracy needed for quantum computers to realise their full potential. The field has seen promising progress with demonstrations of early QEC and real-time decoded experiments. As quantum computers advance towards demonstrating a universal fault-tolerant logical gate set, implementing scalable and low-latency real-time decoding will be crucial to prevent the backlog problem, avoiding an exponential slowdown and maintaining a fast logical clock rate. Here, we demonstrate low-latency feedback with a scalable FPGA decoder integrated into the control system of a superconducting quantum processor. We perform an 8-qubit stability experiment with up to $25$ decoding rounds and a mean decoding time per round below $1$ ${\mu}s$, showing that we avoid the backlog problem even on superconducting hardware with the strictest speed requirements. We observe logical error suppression as the number of decoding rounds is increased. We also implement and time a fast-feedback experiment demonstrating a decoding response time of $9.6$ ${\mu}s$ for a total of $9$ measurement rounds. The decoder throughput and latency developed in this work, combined with continued device improvements, unlock the next generation of experiments that go beyond purely keeping logical qubits alive and into demonstrating building blocks of fault-tolerant computation, such as lattice surgery and magic state teleportation., Comment: 11 pages, 4 figures, Supplementary Information

Data availability across domains often follows a long-tail distribution: a few domains have abundant data, while most face dat . a scarcity. This imbalance poses challenges in training language models uniformly across all domains. In our study, we focus on multilingual settings, where data sizes vary significantly between high- and low-resource languages. Common strategies to address this include upsampling low-resource languages (Temperature Sampling) or upweighting their loss (Scalarization). Although often considered equivalent, this assumption has not been proven, which motivates our study. Through both theoretical and empirical analysis, we identify the conditions under which these approaches are equivalent and when they diverge. Specifically, we demonstrate that these two methods are equivalent under full gradient descent, but this equivalence breaks down with stochastic gradient descent. Empirically, we observe that Temperature Sampling converges more quickly but is prone to overfitting. We argue that this faster convergence is likely due to the lower variance in gradient estimations, as shown theoretically. Based on these insights, we propose Cooldown, a strategy that reduces sampling temperature during training, accelerating convergence without overfitting to low-resource languages. Our method is competitive with existing data re-weighting and offers computational efficiency., Comment: 18 pages

Large language models (LLMs) have achieved remarkable success across various NLP tasks, yet their focus has predominantly been on English due to English-centric pre-training and limited multilingual data. While some multilingual LLMs claim to support for hundreds of languages, models often fail to provide high-quality response for mid- and low-resource languages, leading to imbalanced performance heavily skewed in favor of high-resource languages like English and Chinese. In this paper, we prioritize quality over scaling number of languages, with a focus on multilingual machine translation task, and introduce X-ALMA, a model designed with a commitment to ensuring top-tier performance across 50 diverse languages, regardless of their resource levels. X-ALMA surpasses state-of-the-art open-source multilingual LLMs, such as Aya-101 and Aya-23, in every single translation direction on the FLORES and WMT'23 test datasets according to COMET-22. This is achieved by plug-and-play language-specific module architecture to prevent language conflicts during training and a carefully designed training regimen with novel optimization methods to maximize the translation performance. At the final stage of training regimen, our proposed Adaptive Rejection Preference Optimization (ARPO) surpasses existing preference optimization methods in translation tasks.

This is the preliminary ranking of WMT24 General MT systems based on automatic metrics. The official ranking will be a human evaluation, which is superior to the automatic ranking and supersedes it. The purpose of this report is not to interpret any findings but only provide preliminary results to the participants of the General MT task that may be useful during the writing of the system submission.

With Retrieval Augmented Generation (RAG), Large Language Models (LLMs) are playing a pivotal role in information search and are being adopted globally. Although the multilingual capability of LLMs offers new opportunities to bridge the language barrier, do these capabilities translate into real-life scenarios where linguistic divide and knowledge conflicts between multilingual sources are known occurrences? In this paper, we studied LLM's linguistic preference in a RAG-based information search setting. We found that LLMs displayed systemic bias towards information in the same language as the query language in both information retrieval and answer generation. Furthermore, in scenarios where there is little information in the language of the query, LLMs prefer documents in high-resource languages, reinforcing the dominant views. Such bias exists for both factual and opinion-based queries. Our results highlight the linguistic divide within multilingual LLMs in information search systems. The seemingly beneficial multilingual capability of LLMs may backfire on information parity by reinforcing language-specific information cocoons or filter bubbles further marginalizing low-resource views.

Scalable oversight protocols aim to enable humans to accurately supervise superhuman AI. In this paper we study debate, where two AI's compete to convince a judge; consultancy, where a single AI tries to convince a judge that asks questions; and compare to a baseline of direct question-answering, where the judge just answers outright without the AI. We use large language models (LLMs) as both AI agents and as stand-ins for human judges, taking the judge models to be weaker than agent models. We benchmark on a diverse range of asymmetries between judges and agents, extending previous work on a single extractive QA task with information asymmetry, to also include mathematics, coding, logic and multimodal reasoning asymmetries. We find that debate outperforms consultancy across all tasks when the consultant is randomly assigned to argue for the correct/incorrect answer. Comparing debate to direct question answering, the results depend on the type of task: in extractive QA tasks with information asymmetry debate outperforms direct question answering, but in other tasks without information asymmetry the results are mixed. Previous work assigned debaters/consultants an answer to argue for. When we allow them to instead choose which answer to argue for, we find judges are less frequently convinced by the wrong answer in debate than in consultancy. Further, we find that stronger debater models increase judge accuracy, though more modestly than in previous studies., Comment: 15 pages (53 including appendices). V2: minor correction to Figure 3; add Figure A.9 comparing open vs assigned consultancy; add a reference

While large language models exhibit certain cross-lingual generalization capabilities, they suffer from performance degradation (PD) on unseen closely-related languages (CRLs) and dialects relative to their high-resource language neighbour (HRLN). However, we currently lack a fundamental understanding of what kinds of linguistic distances contribute to PD, and to what extent. Furthermore, studies of cross-lingual generalization are confounded by unknown quantities of CRL language traces in the training data, and by the frequent lack of availability of evaluation data in lower-resource related languages and dialects. To address these issues, we model phonological, morphological, and lexical distance as Bayesian noise processes to synthesize artificial languages that are controllably distant from the HRLN. We analyse PD as a function of underlying noise parameters, offering insights on model robustness to isolated and composed linguistic phenomena, and the impact of task and HRL characteristics on PD. We calculate parameter posteriors on real CRL-HRLN pair data and show that they follow computed trends of artificial languages, demonstrating the viability of our noisers. Our framework offers a cheap solution to estimating task performance on an unseen CRL given HRLN performance using its posteriors, as well as for diagnosing observed PD on a CRL in terms of its linguistic distances from its HRLN, and opens doors to principled methods of mitigating performance degradation., Comment: 9 pages

A majority of language technologies are tailored for a small number of high-resource languages, while relatively many low-resource languages are neglected. One such group, Creole languages, have long been marginalized in academic study, though their speakers could benefit from machine translation (MT). These languages are predominantly used in much of Latin America, Africa and the Caribbean. We present the largest cumulative dataset to date for Creole language MT, including 14.5M unique Creole sentences with parallel translations -- 11.6M of which we release publicly, and the largest bitexts gathered to date for 41 languages -- the first ever for 21. In addition, we provide MT models supporting all 41 Creole languages in 172 translation directions. Given our diverse dataset, we produce a model for Creole language MT exposed to more genre diversity than ever before, which outperforms a genre-specific Creole MT model on its own benchmark for 26 of 34 translation directions., Comment: NAACL 2024

This paper focuses on the opportunities and the ethical and societal risks posed by advanced AI assistants. We define advanced AI assistants as artificial agents with natural language interfaces, whose function is to plan and execute sequences of actions on behalf of a user, across one or more domains, in line with the user's expectations. The paper starts by considering the technology itself, providing an overview of AI assistants, their technical foundations and potential range of applications. It then explores questions around AI value alignment, well-being, safety and malicious uses. Extending the circle of inquiry further, we next consider the relationship between advanced AI assistants and individual users in more detail, exploring topics such as manipulation and persuasion, anthropomorphism, appropriate relationships, trust and privacy. With this analysis in place, we consider the deployment of advanced assistants at a societal scale, focusing on cooperation, equity and access, misinformation, economic impact, the environment and how best to evaluate advanced AI assistants. Finally, we conclude by providing a range of recommendations for researchers, developers, policymakers and public stakeholders.

Recent generative AI systems have demonstrated more advanced persuasive capabilities and are increasingly permeating areas of life where they can influence decision-making. Generative AI presents a new risk profile of persuasion due the opportunity for reciprocal exchange and prolonged interactions. This has led to growing concerns about harms from AI persuasion and how they can be mitigated, highlighting the need for a systematic study of AI persuasion. The current definitions of AI persuasion are unclear and related harms are insufficiently studied. Existing harm mitigation approaches prioritise harms from the outcome of persuasion over harms from the process of persuasion. In this paper, we lay the groundwork for the systematic study of AI persuasion. We first put forward definitions of persuasive generative AI. We distinguish between rationally persuasive generative AI, which relies on providing relevant facts, sound reasoning, or other forms of trustworthy evidence, and manipulative generative AI, which relies on taking advantage of cognitive biases and heuristics or misrepresenting information. We also put forward a map of harms from AI persuasion, including definitions and examples of economic, physical, environmental, psychological, sociocultural, political, privacy, and autonomy harm. We then introduce a map of mechanisms that contribute to harmful persuasion. Lastly, we provide an overview of approaches that can be used to mitigate against process harms of persuasion, including prompt engineering for manipulation classification and red teaming. Future work will operationalise these mitigations and study the interaction between different types of mechanisms of persuasion.

We implement individual addressing and readout of ions in a rigidly rotating planar crystal in a compact, permanent magnet Penning trap. The crystal of $^{40}$Ca$^+$ is trapped and stabilized without defects via a rotating triangular potential. The trapped ion fluorescence is detected in the rotating frame for parallel readout. The qubit is encoded in the metastable D$_{5/2}$ manifold enabling the use of high-power near-infrared laser systems for qubit operations. Addressed $\sigma_z$ operations are realized with a focused AC Stark shifting laser beam. We demonstrate addressing of ions near the center of the crystal and at large radii. Simulations show that the current addressing operation fidelity is limited to $\sim 97\%$ by the ion's thermal extent for the in-plane modes near the Doppler limit, but this could be improved to infidelities $<10^{-3}$ with sub-Doppler cooling. The techniques demonstrated in this paper complete the set of operations for quantum simulation with the platform., Comment: 5 pages, 6 figures

Image retrieval, i.e., finding desired images given a reference image, inherently encompasses rich, multi-faceted search intents that are difficult to capture solely using image-based measures. Recent works leverage text instructions to allow users to more freely express their search intents. However, they primarily focus on image pairs that are visually similar and/or can be characterized by a small set of pre-defined relations. The core thesis of this paper is that text instructions can enable retrieving images with richer relations beyond visual similarity. To show this, we introduce MagicLens, a series of self-supervised image retrieval models that support open-ended instructions. MagicLens is built on a key novel insight: image pairs that naturally occur on the same web pages contain a wide range of implicit relations (e.g., inside view of), and we can bring those implicit relations explicit by synthesizing instructions via foundation models. Trained on 36.7M (query image, instruction, target image) triplets with rich semantic relations mined from the web, MagicLens achieves results comparable with or better than prior best on eight benchmarks of various image retrieval tasks, while maintaining high parameter efficiency with a significantly smaller model size. Additional human analyses on a 1.4M-image unseen corpus further demonstrate the diversity of search intents supported by MagicLens. Code and models are publicly available at https://open-vision-language.github.io/MagicLens/., Comment: ICML 2024 (Oral); Project Website: https://open-vision-language.github.io/MagicLens/

Following natural language instructions by executing actions in digital environments (e.g. web-browsers and REST APIs) is a challenging task for language model (LM) agents. Unfortunately, LM agents often fail to generalize to new environments without human demonstrations. This work presents BAGEL, a method for bootstrapping LM agents without human supervision. BAGEL converts a seed set of randomly explored trajectories or synthetic instructions, into demonstrations, via round-trips between two noisy LM components: an LM labeler which converts a trajectory into a synthetic instruction, and a zero-shot LM agent which maps the synthetic instruction into a refined trajectory. By performing these round-trips iteratively, BAGEL quickly converts the initial distribution of trajectories towards those that are well-described by natural language. We use BAGEL demonstrations to adapt a zero shot LM agent at test time via in-context learning over retrieved demonstrations, and find improvements of over 2-13% absolute on ToolQA and MiniWob++, with up to 13x reduction in execution failures., Comment: ICML 2024 Camera ready version

Moderate-sized large language models (LLMs) -- those with 7B or 13B parameters -- exhibit promising machine translation (MT) performance. However, even the top-performing 13B LLM-based translation models, like ALMA, does not match the performance of state-of-the-art conventional encoder-decoder translation models or larger-scale LLMs such as GPT-4. In this study, we bridge this performance gap. We first assess the shortcomings of supervised fine-tuning for LLMs in the MT task, emphasizing the quality issues present in the reference data, despite being human-generated. Then, in contrast to SFT which mimics reference translations, we introduce Contrastive Preference Optimization (CPO), a novel approach that trains models to avoid generating adequate but not perfect translations. Applying CPO to ALMA models with only 22K parallel sentences and 12M parameters yields significant improvements. The resulting model, called ALMA-R, can match or exceed the performance of the WMT competition winners and GPT-4 on WMT'21, WMT'22 and WMT'23 test datasets., Comment: Accepted at ICML 2024

We show that existing unsupervised methods on large language model (LLM) activations do not discover knowledge -- instead they seem to discover whatever feature of the activations is most prominent. The idea behind unsupervised knowledge elicitation is that knowledge satisfies a consistency structure, which can be used to discover knowledge. We first prove theoretically that arbitrary features (not just knowledge) satisfy the consistency structure of a particular leading unsupervised knowledge-elicitation method, contrast-consistent search (Burns et al. - arXiv:2212.03827). We then present a series of experiments showing settings in which unsupervised methods result in classifiers that do not predict knowledge, but instead predict a different prominent feature. We conclude that existing unsupervised methods for discovering latent knowledge are insufficient, and we contribute sanity checks to apply to evaluating future knowledge elicitation methods. Conceptually, we hypothesise that the identification issues explored here, e.g. distinguishing a model's knowledge from that of a simulated character's, will persist for future unsupervised methods., Comment: 12 pages (38 including references and appendices). First three authors equal contribution, randomised order

Large language models trained primarily in a monolingual setting have demonstrated their ability to generalize to machine translation using zero- and few-shot examples with in-context learning. However, even though zero-shot translations are relatively good, there remains a discernible gap comparing their performance with the few-shot setting. In this paper, we investigate the factors contributing to this gap and find that this gap can largely be closed (for about 70%) by matching the writing styles of the target corpus. Additionally, we explore potential approaches to enhance zero-shot baselines without the need for parallel demonstration examples, providing valuable insights into how these methods contribute to improving translation metrics.

Self-supervised learning (SSL) methods have shown promise for medical imaging applications by learning meaningful visual representations, even when the amount of labeled data is limited. Here, we extend state-of-the-art contrastive learning SSL methods to 2D+time medical ultrasound video data by introducing a modified encoder and augmentation method capable of learning meaningful spatio-temporal representations, without requiring constraints on the input data. We evaluate our method on the challenging clinical task of identifying lung consolidations (an important pathological feature) in ultrasound videos. Using a multi-center dataset of over 27k lung ultrasound videos acquired from over 500 patients, we show that our method can significantly improve performance on downstream localization and classification of lung consolidation. Comparisons against baseline models trained without SSL show that the proposed methods are particularly advantageous when the size of labeled training data is limited (e.g., as little as 5% of the training set)., Comment: ISBI 2023, 2023 IEEE 20th International Symposium on Biomedical Imaging (ISBI)

Text generation models are notoriously vulnerable to errors in the training data. With the wide-spread availability of massive amounts of web-crawled data becoming more commonplace, how can we enhance the robustness of models trained on a massive amount of noisy web-crawled text? In our work, we propose Error Norm Truncation (ENT), a robust enhancement method to the standard training objective that truncates noisy data. Compared to methods that only uses the negative log-likelihood loss to estimate data quality, our method provides a more accurate estimation by considering the distribution of non-target tokens, which is often overlooked by previous work. Through comprehensive experiments across language modeling, machine translation, and text summarization, we show that equipping text generation models with ENT improves generation quality over standard training and previous soft and hard truncation methods. Furthermore, we show that our method improves the robustness of models against two of the most detrimental types of noise in machine translation, resulting in an increase of more than 2 BLEU points over the MLE baseline when up to 50% of noise is added to the data., Comment: ICLR 2024

To unleash the potential of quantum computers, noise effects on qubits' performance must be carefully managed. The decoders responsible for diagnosing noise-induced computational errors must use resources efficiently to enable scaling to large qubit counts and cryogenic operation. Additionally, they must operate at speed, to avoid an exponential slowdown in the logical clock rate of the quantum computer. To overcome such challenges, we introduce the Collision Clustering decoder and implement it on FPGA and ASIC hardware. We simulate logical memory experiments using the leading quantum error correction scheme, the surface code, and demonstrate MHz decoding speed - matching the requirements of fast-operating modalities such as superconducting qubits - up to an 881 and 1057 qubits surface code with the FPGA and ASIC, respectively. The ASIC design occupies 0.06 mm$^2$ and consumes only 8 mW of power. Our decoder is both highly performant and resource efficient, unlocking a viable path to practically realising fault-tolerant quantum computers., Comment: 13 pages, 7 figures

An extensive library of symptom inventories has been developed over time to measure clinical symptoms, but this variety has led to several long standing issues. Most notably, results drawn from different settings and studies are not comparable, which limits reproducibility. Here, we present an artificial intelligence (AI) approach using semantic textual similarity (STS) to link symptoms and scores across previously incongruous symptom inventories. We tested the ability of four pre-trained STS models to screen thousands of symptom description pairs for related content - a challenging task typically requiring expert panels. Models were tasked to predict symptom severity across four different inventories for 6,607 participants drawn from 16 international data sources. The STS approach achieved 74.8% accuracy across five tasks, outperforming other models tested. This work suggests that incorporating contextual, semantic information can assist expert decision-making processes, yielding gains for both general and disease-specific clinical assessment.

The lack of quality labeled data is one of the main bottlenecks for training Deep Learning models. As the task increases in complexity, there is a higher penalty for overfitting and unstable learning. The typical paradigm employed today is Self-Supervised learning, where the model attempts to learn from a large corpus of unstructured and unlabeled data and then transfer that knowledge to the required task. Some notable examples of self-supervision in other modalities are BERT for Large Language Models, Wav2Vec for Speech Recognition, and the Masked AutoEncoder for Vision, which all utilize Transformers to solve a masked prediction task. GeoAI is uniquely poised to take advantage of the self-supervised methodology due to the decades of data collected, little of which is precisely and dependably annotated. Our goal is to extract building and road segmentations from Digital Elevation Models (DEM) that provide a detailed topography of the earths surface. The proposed architecture is the Masked Autoencoder pre-trained on ImageNet (with the limitation that there is a large domain discrepancy between ImageNet and DEM) with an UperNet Head for decoding segmentations. We tested this model with 450 and 50 training images only, utilizing roughly 5% and 0.5% of the original data respectively. On the building segmentation task, this model obtains an 82.1% Intersection over Union (IoU) with 450 Images and 69.1% IoU with only 50 images. On the more challenging road detection task the model obtains an 82.7% IoU with 450 images and 73.2% IoU with only 50 images. Any hand-labeled dataset made today about the earths surface will be immediately obsolete due to the constantly changing nature of the landscape. This motivates the clear necessity for data-efficient learners that can be used for a wide variety of downstream tasks.

One of the most surprising puzzles in neural network generalisation is grokking: a network with perfect training accuracy but poor generalisation will, upon further training, transition to perfect generalisation. We propose that grokking occurs when the task admits a generalising solution and a memorising solution, where the generalising solution is slower to learn but more efficient, producing larger logits with the same parameter norm. We hypothesise that memorising circuits become more inefficient with larger training datasets while generalising circuits do not, suggesting there is a critical dataset size at which memorisation and generalisation are equally efficient. We make and confirm four novel predictions about grokking, providing significant evidence in favour of our explanation. Most strikingly, we demonstrate two novel and surprising behaviours: ungrokking, in which a network regresses from perfect to low test accuracy, and semi-grokking, in which a network shows delayed generalisation to partial rather than perfect test accuracy.

The trapped-ion quantum charge-coupled device (QCCD) architecture is a leading candidate for advanced quantum information processing. In current QCCD implementations, imperfect ion transport and anomalous heating can excite ion motion during a calculation. To counteract this, intermediate cooling is necessary to maintain high-fidelity gate performance. Cooling the computational ions sympathetically with ions of another species, a commonly employed strategy, creates a significant runtime bottleneck. Here, we demonstrate a different approach we call exchange cooling. Unlike sympathetic cooling, exchange cooling does not require trapping two different atomic species. The protocol introduces a bank of "coolant" ions which are repeatedly laser cooled. A computational ion can then be cooled by transporting a coolant ion into its proximity. We test this concept experimentally with two $^{40}\mathrm{Ca}^{+}$ ions, executing the necessary transport in 107 $\mathrm{\mu s}$, an order of magnitude faster than typical sympathetic cooling durations. We remove over 96%, and as many as 102(5) quanta, of axial motional energy from the computational ion. We verify that re-cooling the coolant ion does not decohere the computational ion. This approach validates the feasibility of a single-species QCCD processor, capable of fast quantum simulation and computation., Comment: 22 pages, 7 figures; matching publication

This paper concerns the relationships between continued fractions and the geometry of the Stern-Brocot diagram. Each rational number can be expressed as a continued fraction $[a_0; a_1, \ldots, a_n]$ whose terms $a_i$ are integers and are positive if $i \geq 1$. Select an index $i \in \{ 1, \ldots, n \}$ and replace $a_i$ with an integer $m$ to obtain a continued fraction expansion for an extended rational $\alpha_m \in \mathbb{Q} \cup \{ \infty \}$. This paper shows that the vertices of the Stern-Brocot diagram corresponding to the numbers $\{ \alpha_m \}_{m \in \mathbb{Z}}$ lie on a pair of (extended) Euclidean lines across the diagram. The slopes of these two lines differ only by a sign change and they meet at the point $L=\left([a_0; a_1, \ldots, a_{i-1}], 0\right) \in \mathbb{R}^2$. Moreover, as $\lvert m \rvert \to \infty$, the associated vertices move down these lines and converge to $L$. This paper concludes with a discussion which interprets this result in the context of 2-bridge link complements and Thurston's work on hyperbolic Dehn surgery., Comment: 11 pages, comments welcome

Frame-by-frame annotation of bounding boxes by clinical experts is often required to train fully supervised object detection models on medical video data. We propose a method for improving object detection in medical videos through weak supervision from video-level labels. More concretely, we aggregate individual detection predictions into video-level predictions and extend a teacher-student training strategy to provide additional supervision via a video-level loss. We also introduce improvements to the underlying teacher-student framework, including methods to improve the quality of pseudo-labels based on weak supervision and adaptive schemes to optimize knowledge transfer between the student and teacher networks. We apply this approach to the clinically important task of detecting lung consolidations (seen in respiratory infections such as COVID-19 pneumonia) in medical ultrasound videos. Experiments reveal that our framework improves detection accuracy and robustness compared to baseline semi-supervised models, and improves efficiency in data and annotation usage., Comment: IPMI 2023

To foster the development of new models for collaborative AI-assisted report generation, we introduce MegaWika, consisting of 13 million Wikipedia articles in 50 diverse languages, along with their 71 million referenced source materials. We process this dataset for a myriad of applications, going beyond the initial Wikipedia citation extraction and web scraping of content, including translating non-English articles for cross-lingual applications and providing FrameNet parses for automated semantic analysis. MegaWika is the largest resource for sentence-level report generation and the only report generation dataset that is multilingual. We manually analyze the quality of this resource through a semantically stratified sample. Finally, we provide baseline results and trained models for crucial steps in automated report generation: cross-lingual question answering and citation retrieval., Comment: Submitted to ACL, 2023

Much of the previous work towards digital agents for graphical user interfaces (GUIs) has relied on text-based representations (derived from HTML or other structured data sources), which are not always readily available. These input representations have been often coupled with custom, task-specific action spaces. This paper focuses on creating agents that interact with the digital world using the same conceptual interface that humans commonly use -- via pixel-based screenshots and a generic action space corresponding to keyboard and mouse actions. Building upon recent progress in pixel-based pretraining, we show, for the first time, that it is possible for such agents to outperform human crowdworkers on the MiniWob++ benchmark of GUI-based instruction following tasks.

We present the training recipe and results of scaling up PaLI-X, a multilingual vision and language model, both in terms of size of the components and the breadth of its training task mixture. Our model achieves new levels of performance on a wide-range of varied and complex tasks, including multiple image-based captioning and question-answering tasks, image-based document understanding and few-shot (in-context) learning, as well as object detection, video question answering, and video captioning. PaLI-X advances the state-of-the-art on most vision-and-language benchmarks considered (25+ of them). Finally, we observe emerging capabilities, such as complex counting and multilingual object detection, tasks that are not explicitly in the training mix.

Zero-shot cross-lingual transfer is when a multilingual model is trained to perform a task in one language and then is applied to another language. Although the zero-shot cross-lingual transfer approach has achieved success in various classification tasks, its performance on natural language generation tasks falls short in quality and sometimes outputs an incorrect language. In our study, we show that the fine-tuning process learns language invariant representations, which is beneficial for classification tasks but harmful for generation tasks. Motivated by this, we propose a simple method to regularize the model from learning language invariant representations and a method to select model checkpoints without a development set in the target language, both resulting in better generation quality. Experiments on three semantically diverse generation tasks show that our method reduces the accidental translation problem by 68% and improves the ROUGE-L score by 1.5 on average., Comment: Findings of ACL 2023

The mixed-size placement benchmarks described in this book chapter directly address several of the shortcomings in previously published suboptimality benchmarks. Two new sets of placement examples are constructed, one targeting the role of nonlocal nets in suboptimality, and another targeting the role of white space and large variations in module sizes. The first set, PEKO-MC, is a set of standard-cell circuits with nonlocal nets in known optimal placements. A given netlist is modified so as to render a given placement for it optimal for the new netlist. Cell dimensions and locations are not changed, net-degree statistics are matched exactly, and over 60% of the original netlist is left unchanged. The second set, PEKO-MS, incorporates a parametrized percentage of white space into a mixed-size placement which precisely matches given macro dimensions and locations as well as the net-degree distributions of the ISPD 2005 benchmark suite [21]. HPWL for the placements generated for the PEKO-MS circuits are proven to be less than 3% above optimal for most cases and within 8% of optimal on all cases., Comment: Version of Record: https://link.springer.com/book/10.1007/978-0-387-68739-1 Pre-print of Chapter 2 from the following work: Gi-Joon Nam and Jason Cong, Modern Circuit Placement Best Practices and Results, 2007, Springer, reproduced with permission of Springer Science+Business Media, LLC. The final authenticated version is available online at: http://dx.doi.org/10.1007/978-0-387-68739-1

Incorporating language-specific (LS) modules is a proven method to boost performance in multilingual machine translation. This approach bears similarity to Mixture-of-Experts (MoE) because it does not inflate FLOPs. However, the scalability of this approach to hundreds of languages (experts) tends to be unmanageable due to the prohibitive number of parameters introduced by full-rank matrices in fully-connected layers. In this work, we introduce the Language-Specific Matrix Synthesis (LMS) method. This approach constructs LS modules by generating low-rank matrices from two significantly smaller matrices to approximate the full-rank matrix. Furthermore, we condense multilingual knowledge from multiple LS modules into a single shared module with the Fuse Distillation (FD) technique to improve the efficiency of inference and model serialization. We show that our LMS method significantly outperforms previous LS methods and MoE methods with the same amount of extra parameters, e.g., 1.73 BLEU points over the Switch Transformer on many-to-many multilingual machine translation. Importantly, LMS is able to have comparable translation performance with much fewer parameters., Comment: Accepted at the main conference of EMNLP 2023