This paper initiates the study of active learning for exact recovery of partitions exclusively through access to a same-cluster oracle in the presence of bounded adversarial error. We first highlight a novel connection between learning partitions and correlation clustering. Then we use this connection to build a R\'enyi-Ulam style analytical framework for this problem, and prove upper and lower bounds on its worst-case query complexity. Further, we bound the expected performance of a relevant randomized algorithm. Finally, we study the relationship between adaptivity and query complexity for this problem and related variants.

In this work, we propose a new class of learnable optimizers, called \textit{Mnemosyne}. It is based on the novel spatio-temporal low-rank implicit attention Transformers that can learn to train entire neural network architectures, including other Transformers, without any task-specific optimizer tuning. We show that Mnemosyne: (a) outperforms popular LSTM optimizers (also with new feature engineering to mitigate catastrophic forgetting of LSTMs), (b) can successfully train Transformers while using simple meta-training strategies that require minimal computational resources, (c) matches accuracy-wise SOTA hand-designed optimizers with carefully tuned hyper-parameters (often producing top performing models). Furthermore, Mnemosyne provides space complexity comparable to that of its hand-designed first-order counterparts, which allows it to scale to training larger sets of parameters. We conduct an extensive empirical evaluation of Mnemosyne on: (a) fine-tuning a wide range of Vision Transformers (ViTs) from medium-size architectures to massive ViT-Hs (36 layers, 16 heads), (b) pre-training BERT models and (c) soft prompt-tuning large 11B+ T5XXL models. We complement our results with a comprehensive theoretical analysis of the compact associative memory used by Mnemosyne which we believe was never done before.

Mis- and disinformation are a substantial global threat to our security and safety. To cope with the scale of online misinformation, researchers have been working on automating fact-checking by retrieving and verifying against relevant evidence. However, despite many advances, a comprehensive evaluation of the possible attack vectors against such systems is still lacking. Particularly, the automated fact-verification process might be vulnerable to the exact disinformation campaigns it is trying to combat. In this work, we assume an adversary that automatically tampers with the online evidence in order to disrupt the fact-checking model via camouflaging the relevant evidence or planting a misleading one. We first propose an exploratory taxonomy that spans these two targets and the different threat model dimensions. Guided by this, we design and propose several potential attack methods. We show that it is possible to subtly modify claim-salient snippets in the evidence and generate diverse and claim-aligned evidence. Thus, we highly degrade the fact-checking performance under many different permutations of the taxonomy's dimensions. The attacks are also robust against post-hoc modifications of the claim. Our analysis further hints at potential limitations in models' inference when faced with contradicting evidence. We emphasize that these attacks can have harmful implications on the inspectable and human-in-the-loop usage scenarios of such models, and we conclude by discussing challenges and directions for future defenses.

Evaluating counterfactual statements is a fundamental problem for many approaches to causal reasoning [40]. Such reasoning can for instance be used to explain erroneous system behavior with a counterfactual statement such as'If the input i at the first position of the observed computation π had not been enabled then the system would not have reached an error e.' which can be formalized using the counterfactual operator and the temporal operator F: π ( i) ( Fe). Since the foundational work by Lewis[38] on the formal semantics of counterfactual conditionals, many applications for counterfactuals [28, 5, 34, 46, 3, 15] and some theoretical results on the decidability of the original theory [37] and related notions [20, 2] have been discovered. Still, certain domains have proven elusive for a long time, for instance, theories involving higher-order reasoning and an infinite number of variables. In this paper, we consider a domain that combines both of these aspects: temporal reasoning over infinite sequences. In particular, we consider counterfactual conditionals that relate two properties expressed in temporal logics, such as the temporal property F e from the introductory example. Temporal logics are used ubiquitously as high-level specifications for verification [21, 4] and synthesis [22, 41], and recently have also found use in specifying reinforcement learning tasks [32, 39]. Our work lifts the language of counterfactual reasoning to similar high-level expressions.

Real-world deployment of computer vision systems, including in the discovery processes of biomedical research, requires causal representations that are invariant to contextual nuisances and generalize to new data. Leveraging the internal replicate structure of two novel single-cell fluorescent microscopy datasets, we propose generally applicable tests to assess the extent to which models learn causal representations across increasingly challenging levels of OOD-generalization. We show that despite seemingly strong performance, as assessed by other established metrics, both naive and contemporary baselines designed to ward against confounding, collapse on these tests. We introduce a new method, Interventional Style Transfer (IST), that substantially improves OOD generalization by generating interventional training distributions in which spurious correlations between biological causes and nuisances are mitigated. We publish our code and datasets.

This paper investigates the effect of tokenizers on the downstream performance of pretrained language models (PLMs) in scriptio continua languages where no explicit spaces exist between words, using Japanese as a case study. The tokenizer for such languages often consists of a morphological analyzer and a subword tokenizer, requiring us to conduct a comprehensive study of all possible pairs. However, previous studies lack this comprehensiveness. We therefore train extensive sets of tokenizers, build a PLM using each, and measure the downstream performance on a wide range of tasks. Our results demonstrate that each downstream task has a different optimal morphological analyzer, and that it is better to use Byte-Pair-Encoding or Unigram rather than WordPiece as a subword tokenizer, regardless of the type of task.

Despite several solutions and experiments have been conducted recently addressing image super-resolution (SR), boosted by deep learning (DL) techniques, they do not usually design evaluations with high scaling factors, capping it at 2x or 4x. Moreover, the datasets are generally benchmarks which do not truly encompass significant diversity of domains to proper evaluate the techniques. It is also interesting to remark that blind SR is attractive for real-world scenarios since it is based on the idea that the degradation process is unknown, and hence techniques in this context rely basically on low-resolution (LR) images. In this article, we present a high-scale (8x) controlled experiment which evaluates five recent DL techniques tailored for blind image SR: Adaptive Pseudo Augmentation (APA), Blind Image SR with Spatially Variant Degradations (BlindSR), Deep Alternating Network (DAN), FastGAN, and Mixture of Experts Super-Resolution (MoESR). We consider 14 small datasets from five different broader domains which are: aerial, fauna, flora, medical, and satellite. Another distinctive characteristic of our evaluation is that some of the DL approaches were designed for single-image SR but others not. Two no-reference metrics were selected, being the classical natural image quality evaluator (NIQE) and the recent transformer-based multi-dimension attention network for no-reference image quality assessment (MANIQA) score, to assess the techniques. Overall, MoESR can be regarded as the best solution although the perceptual quality of the created HR images of all the techniques still needs to improve. Supporting code: https://github.com/vsantjr/DL_BlindSR. Datasets: https://www.kaggle.com/datasets/valdivinosantiago/dl-blindsr-datasets.

We describe some recent approaches to score-based explanations for query answers in databases. The focus is on work done by the author and collaborators. Special emphasis is placed on the use of counterfactual reasoning for score specification and computation. Several examples that illustrate the flexibility of these methods are shown.

In schema-guided dialogue state tracking models estimate the current state of a conversation using natural language descriptions of the service schema for generalization to unseen services. Prior generative approaches which decode slot values sequentially do not generalize well to variations in schema, while discriminative approaches separately encode history and schema and fail to account for inter-slot and intent-slot dependencies. We introduce SPLAT, a novel architecture which achieves better generalization and efficiency than prior approaches by constraining outputs to a limited prediction space. At the same time, our model allows for rich attention among descriptions and history while keeping computation costs constrained by incorporating linear-time attention. We demonstrate the effectiveness of our model on the Schema-Guided Dialogue (SGD) and MultiWOZ datasets. Our approach significantly improves upon existing models achieving 85.3 JGA on the SGD dataset. Further, we show increased robustness on the SGD-X benchmark: our model outperforms the more than 30$\times$ larger D3ST-XXL model by 5.0 points.

The wide applicability and adaptability of generative large language models (LLMs) has enabled their rapid adoption. While the pre-trained models can perform many tasks, such models are often fine-tuned to improve their performance on various downstream applications. However, this leads to issues over violation of model licenses, model theft, and copyright infringement. Moreover, recent advances show that generative technology is capable of producing harmful content which exacerbates the problems of accountability within model supply chains. Thus, we need a method to investigate how a model was trained or a piece of text was generated and what their pre-trained base model was. In this paper we take the first step to address this open problem by tracing back the origin of a given fine-tuned LLM to its corresponding pre-trained base model. We consider different knowledge levels and attribution strategies, and find that we can correctly trace back 8 out of the 10 fine tuned models with our best method.