Committee

arXiv:2505.10507v1 Announce Type: new Abstract: Translation-based strategies for cross-lingual transfer XLT such as translate-train — training on noisy target language data translated from the source language — and translate-test — evaluating on noisy source language data translated from the target language — are competitive XLT baselines. In XLT for token classification tasks, however, these strategies include label projection, the challenging step of mapping the labels from each token in the original sentence to its counterpart(s) in the translation. Although word aligners (WAs) are commonly used for label projection, the low-level design decisions for applying them to translation-based XLT have not been systematically investigated. Moreover, recent marker-based methods, which project labeled spans by inserting tags around them before (or after) translation, claim to outperform WAs in label projection for XLT. In this work, we revisit WAs for label projection, systematically investigating the effects of low-level design decisions on token-level XLT: (i) the algorithm for projecting labels between (multi-)token spans, (ii) filtering strategies to reduce the number of noisily mapped labels, and (iii) the pre-tokenization of the translated sentences. We find that all of these substantially impact translation-based XLT performance and show that, with optimized choices, XLT with WA offers performance at least comparable to that of marker-based methods. We then introduce a new projection strategy that ensembles translate-train and translate-test predictions and demonstrate that it substantially outperforms the marker-based projection. Crucially, we show that our proposed ensembling also reduces sensitivity to low-level WA design choices, resulting in more robust XLT for token classification tasks.

arXiv:2505.09738v1 Announce Type: new Abstract: Pretrained language models (LLMs) are often constrained by their fixed tokenization schemes, leading to inefficiencies and performance limitations, particularly for multilingual or specialized applications. This tokenizer lock-in presents significant challenges. standard methods to overcome this often require prohibitive computational resources. Although tokenizer replacement with heuristic initialization aims to reduce this burden, existing methods often require exhaustive residual fine-tuning and still may not fully preserve semantic nuances or adequately address the underlying compression inefficiencies. Our framework introduces two innovations: first, Tokenadapt, a model-agnostic tokenizer transplantation method, and second, novel pre-tokenization learning for multi-word Supertokens to enhance compression and reduce fragmentation. Tokenadapt initializes new unique token embeddings via a hybrid heuristic that combines two methods: a local estimate based on subword decomposition using the old tokenizer, and a global estimate utilizing the top-k semantically similar tokens from the original vocabulary. This methodology aims to preserve semantics while significantly minimizing retraining requirements. Empirical investigations validate both contributions: the transplantation heuristic successfully initializes unique tokens, markedly outperforming conventional baselines and sophisticated methods including Transtokenizer and ReTok, while our Supertokens achieve notable compression gains. Our zero-shot perplexity results demonstrate that the TokenAdapt hybrid initialization consistently yields lower perplexity ratios compared to both ReTok and TransTokenizer baselines across different base models and newly trained target tokenizers. TokenAdapt typically reduced the overall perplexity ratio significantly compared to ReTok, yielding at least a 2-fold improvement in these aggregate scores.

arXiv:2504.17671v3 Announce Type: replace Abstract: This study addresses the critical challenge of hallucination mitigation in Large Vision-Language Models (LVLMs) for Visual Question Answering (VQA) tasks through a Split Conformal Prediction (SCP) framework. While LVLMs excel in multi-modal reasoning, their outputs often exhibit hallucinated content with high confidence, posing risks in safety-critical applications. We propose a model-agnostic uncertainty quantification method that integrates dynamic threshold calibration and cross-modal consistency verification. By partitioning data into calibration and test sets, the framework computes nonconformity scores to construct prediction sets with statistical guarantees under user-defined risk levels ($alpha$). Key innovations include: (1) rigorous control of textbf{marginal coverage} to ensure empirical error rates remain strictly below $alpha$; (2) dynamic adjustment of prediction set sizes inversely with $alpha$, filtering low-confidence outputs; (3) elimination of prior distribution assumptions and retraining requirements. Evaluations on benchmarks (ScienceQA, MMMU) with eight LVLMs demonstrate that SCP enforces theoretical guarantees across all $alpha$ values. The framework achieves stable performance across varying calibration-to-test split ratios, underscoring its robustness for real-world deployment in healthcare, autonomous systems, and other safety-sensitive domains. This work bridges the gap between theoretical reliability and practical applicability in multi-modal AI systems, offering a scalable solution for hallucination detection and uncertainty-aware decision-making.

arXiv:2505.05145v2 Announce Type: replace-cross Abstract: To perform in-context learning, language models must extract signals from individual few-shot examples, aggregate these into a learned prediction rule, and then apply this rule to new examples. How is this implemented in the forward pass of modern transformer models? To study this, we consider a structured family of few-shot learning tasks for which the true prediction rule is to add an integer $k$ to the input. We find that Llama-3-8B attains high accuracy on this task for a range of $k$, and localize its few-shot ability to just three attention heads via a novel optimization approach. We further show the extracted signals lie in a six-dimensional subspace, where four of the dimensions track the unit digit and the other two dimensions track overall magnitude. We finally examine how these heads extract information from individual few-shot examples, identifying a self-correction mechanism in which mistakes from earlier examples are suppressed by later examples. Our results demonstrate how tracking low-dimensional subspaces across a forward pass can provide insight into fine-grained computational structures.

arXiv:2505.09724v1 Announce Type: new Abstract: Analyzing texts such as open-ended responses, headlines, or social media posts is a time- and labor-intensive process highly susceptible to bias. LLMs are promising tools for text analysis, using either a predefined (top-down) or a data-driven (bottom-up) taxonomy, without sacrificing quality. Here we present a step-by-step tutorial to efficiently develop, test, and apply taxonomies for analyzing unstructured data through an iterative and collaborative process between researchers and LLMs. Using personal goals provided by participants as an example, we demonstrate how to write prompts to review datasets and generate a taxonomy of life domains, evaluate and refine the taxonomy through prompt and direct modifications, test the taxonomy and assess intercoder agreements, and apply the taxonomy to categorize an entire dataset with high intercoder reliability. We discuss the possibilities and limitations of using LLMs for text analysis.

arXiv:2411.03343v2 Announce Type: replace-cross Abstract: Jailbreaks have been a central focus of research regarding the safety and reliability of large language models (LLMs), yet the mechanisms underlying these attacks remain poorly understood. While previous studies have predominantly relied on linear methods to detect jailbreak attempts and model refusals, we take a different approach by examining both linear and non-linear features in prompts that lead to successful jailbreaks. First, we introduce a novel dataset comprising 10,800 jailbreak attempts spanning 35 diverse attack methods. Leveraging this dataset, we train probes to classify successful from unsuccessful jailbreaks using the latent representations corresponding to prompt tokens. Notably, we find that even when probes achieve high accuracy in predicting the success of jailbreaks, their performance often fails to generalize to unseen attack methods. This reveals that different jailbreaking strategies exploit different non-linear, non-universal features. Next, we demonstrate that non-linear probes provide a powerful tool for steering model behavior. Specifically, we use these probes to guide targeted latent space perturbations, enabling us to effectively modulate the model’s robustness against jailbreaks. Overall, our findings challenge the assumption that jailbreaks can be fully understood through linear or simple universal prompt features alone, highlighting the importance of a nuanced understanding of the mechanisms behind LLM vulnerabilities.

arXiv:2505.08971v1 Announce Type: cross Abstract: In standard large vision-language models (LVLMs) pre-training, the model typically maximizes the joint probability of the caption conditioned on the image via next-token prediction (NTP); however, since only a small subset of caption tokens directly relates to the visual content, this naive NTP unintentionally fits the model to noise and increases the risk of hallucination. We present PRIOR, a simple vision-language pre-training approach that addresses this issue by prioritizing image-related tokens through differential weighting in the NTP loss, drawing from the importance sampling framework. PRIOR introduces a reference model-a text-only large language model (LLM) trained on the captions without image inputs, to weight each token based on its probability for LVLMs training. Intuitively, tokens that are directly related to the visual inputs are harder to predict without the image and thus receive lower probabilities from the text-only reference LLM. During training, we implement a token-specific re-weighting term based on the importance scores to adjust each token’s loss. We implement PRIOR in two distinct settings: LVLMs with visual encoders and LVLMs without visual encoders. We observe 19% and 8% average relative improvement, respectively, on several vision-language benchmarks compared to NTP. In addition, PRIOR exhibits superior scaling properties, as demonstrated by significantly higher scaling coefficients, indicating greater potential for performance gains compared to NTP given increasing compute and data.