Gerard de Melo

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Ziyi Lin, Shijie Geng, Renrui Zhang et al.

Video recognition has been dominated by the end-to-end learning paradigm -- first initializing a video recognition model with weights of a pretrained image model and then conducting end-to-end training on videos. This enables the video network to benefit from the pretrained image model. However, this requires substantial computation and memory resources for finetuning on videos and the alternative of directly using pretrained image features without finetuning the image backbone leads to subpar results. Fortunately, recent advances in Contrastive Vision-Language Pre-training (CLIP) pave the way for a new route for visual recognition tasks. Pretrained on large open-vocabulary image-text pair data, these models learn powerful visual representations with rich semantics. In this paper, we present Efficient Video Learning (EVL) -- an efficient framework for directly training high-quality video recognition models with frozen CLIP features. Specifically, we employ a lightweight Transformer decoder and learn a query token to dynamically collect frame-level spatial features from the CLIP image encoder. Furthermore, we adopt a local temporal module in each decoder layer to discover temporal clues from adjacent frames and their attention maps. We show that despite being efficient to train with a frozen backbone, our models learn high quality video representations on a variety of video recognition datasets. Code is available at https://github.com/OpenGVLab/efficient-video-recognition.

Zetian Ouyang, Linlin Wang, Gerard de Melo et al.

Despite the pivotal role of numerical reasoning as the cornerstone of mathematical capabilities in large language models (LLMs) across applications, few benchmarks evaluate LLMs by integrating numerical processing and mathematical reasoning, hindering the interpretability of failures in math tasks. We introduce PyraMathBench, a comprehensive hierarchical benchmark with 32,505 questions derived from 7,404 math word problems, spanning 4 key cognitive aspects, 14 subcategories, and 2 modalities. Experiments reveal that LLMs' performance is severely compromised by inadequate numerical computation and weak handling of abstract numerical questions. To address this, we propose the Smart Optimization & Learning-based VErsatile module (SOLVE) and Interactive Relative Policy Optimization (IRPO), which enhance LLMs' numerical-mathematical synergy via efficient tool calls (fuzzy matching and low-quality call rejection). Comparative experiments show Qwen-2.5 achieves a 5.0 score improvement with SOLVE and IRPO training.

Ruangrin Ldallitsakool, Margarita Bugueño, Gerard de Melo

Recent NLP systems commonly represent documents as linear token sequences. Although this captures sequential order, it can hinder modeling long-range dependencies and global document structure, especially for long texts. This paper proposes a data-driven method to automatically construct graph-based document representations. Building upon the recent work of Bugueño and de Melo (2025), we leverage the dynamic sliding-window attention module to effectively capture local and mid-range semantic dependencies between sentences, as well as structural relations within documents. Graph Attention Networks (GATs) trained on our learned graphs achieve competitive results on document classification while requiring lower computational resources than previous approaches. We further present an exploratory evaluation of the proposed graph construction method for extractive document summarization, highlighting both its potential and current limitations. The implementation of this project can be found on GitHub.

Jeff Z. Pan, Simon Razniewski, Jan-Christoph Kalo et al.

Large Language Models (LLMs) have taken Knowledge Representation -- and the world -- by storm. This inflection point marks a shift from explicit knowledge representation to a renewed focus on the hybrid representation of both explicit knowledge and parametric knowledge. In this position paper, we will discuss some of the common debate points within the community on LLMs (parametric knowledge) and Knowledge Graphs (explicit knowledge) and speculate on opportunities and visions that the renewed focus brings, as well as related research topics and challenges.

Sepehr Janghorbani, Gerard de Melo

Recent breakthroughs in self supervised training have led to a new class of pretrained vision language models. While there have been investigations of bias in multimodal models, they have mostly focused on gender and racial bias, giving much less attention to other relevant groups, such as minorities with regard to religion, nationality, sexual orientation, or disabilities. This is mainly due to lack of suitable benchmarks for such groups. We seek to address this gap by providing a visual and textual bias benchmark called MMBias, consisting of around 3,800 images and phrases covering 14 population subgroups. We utilize this dataset to assess bias in several prominent self supervised multimodal models, including CLIP, ALBEF, and ViLT. Our results show that these models demonstrate meaningful bias favoring certain groups. Finally, we introduce a debiasing method designed specifically for such large pre-trained models that can be applied as a post-processing step to mitigate bias, while preserving the remaining accuracy of the model.

Xiang Hu, Haitao Mi, Liang Li et al.

Recently CKY-based models show great potential in unsupervised grammar induction thanks to their human-like encoding paradigm, which runs recursively and hierarchically, but requires $O(n^3)$ time-complexity. Recursive Transformer based on Differentiable Trees (R2D2) makes it possible to scale to large language model pre-training even with complex tree encoder by introducing a heuristic pruning method. However, the rule-based pruning approach suffers from local optimum and slow inference issues. In this paper, we fix those issues in a unified method. We propose to use a top-down parser as a model-based pruning method, which also enables parallel encoding during inference. Typically, our parser casts parsing as a split point scoring task, which first scores all split points for a given sentence, and then recursively splits a span into two by picking a split point with the highest score in the current span. The reverse order of the splits is considered as the order of pruning in R2D2 encoder. Beside the bi-directional language model loss, we also optimize the parser by minimizing the KL distance between tree probabilities from parser and R2D2. Our experiments show that our Fast-R2D2 improves performance significantly in grammar induction and achieves competitive results in downstream classification tasks.

Harshil Darji, Martin Heckelmann, Christina Kratsch et al.

Statutory references are central to legal language understanding, but are difficult to process automatically, as they appear in compact and variable surface forms, may combine multiple targets, use special abbreviations, and often point to lower-level units. Existing tools for German focus either on parsing references from legal documents or accessing statutory text once citations are explicit. This paper introduces bundesrecht, an open resource for German statutory reference processing, consisting of a software library and a structured corpus of German federal law. The library parses, normalizes, and resolves German statutory references, mapping raw citation strings to structured objects, expanding compact references into canonical forms, and linking them to statutory provisions. The accompanying dataset preserves the internal hierarchy of statutes from laws to fine-granular subclauses. We evaluate the parser and normalizer on 2,944 annotated German legal references using strict exact-match and micro information extraction metrics. We further evaluate canonical reference deduplication and show that normalized references group real citation surface variants far more reliably than string matching. bundesrecht is the first open resource that covers German statutory reference processing as an end-to-end pipeline, from raw citation string to resolved statutory provision, and is available on PyPI.

Johannes Hagemann, Samuel Weinbach, Konstantin Dobler et al.

Efficiently training large language models requires parallelizing across hundreds of hardware accelerators and invoking various compute and memory optimizations. When combined, many of these strategies have complex interactions regarding the final training efficiency. Prior work tackling this problem did not have access to the latest set of optimizations, such as FlashAttention or sequence parallelism. In this work, we conduct a comprehensive ablation study of possible training configurations for large language models. We distill this large study into several key recommendations for the most efficient training. For instance, we find that using a micro-batch size of 1 usually enables the most efficient training layouts. Larger micro-batch sizes necessitate activation checkpointing or higher degrees of model parallelism and also lead to larger pipeline bubbles. Our most efficient configurations enable us to achieve state-of-the-art training efficiency results over a range of model sizes, most notably a Model FLOPs utilization of 70.5% when training a Llama 13B model.

Weixing Wang, Liudvikas Zekas, Anton Hackl et al.

Unified Multimodal Models (uMMs) aim to support both visual understanding and visual generation within a shared representation. However, existing evaluation protocols assess these two capabilities independently and do not examine whether they are semantically aligned. As a result, it remains unclear whether current uMMs learn coherent unified representations that remain consistent across tasks given a visual concept. We introduce XTC-Bench, a scene-graph-grounded evaluation framework that measures cross-task visual semantic consistency. By deriving both generation prompts and understanding queries from a structured scene graph, our framework enables fact-level alignment analysis across objects, attributes, and relations. We propose Continuous Cross-Task Agreement (CCTA), a fine-grained metric that quantifies semantic agreement between generation and understanding over matched atomic facts, isolating internal consistency from standalone task accuracy. Extensive experiments on eight open-source and one commercial unified models reveal that high generation or understanding performance does not imply strong cross-task alignment, and architectural analysis shows consistency is governed by how tightly learning objectives are coupled across modalities, not by architectural unification alone. XTC-Bench provides a reproducible and model-agnostic framework for diagnosing representation-level misalignment, offering a concrete direction for advancing unified multimodal modeling beyond isolated task performance.

Terry Yue Zhuo, Yaqing Liao, Yuecheng Lei et al.

We introduce ViLPAct, a novel vision-language benchmark for human activity planning. It is designed for a task where embodied AI agents can reason and forecast future actions of humans based on video clips about their initial activities and intents in text. The dataset consists of 2.9k videos from \charades extended with intents via crowdsourcing, a multi-choice question test set, and four strong baselines. One of the baselines implements a neurosymbolic approach based on a multi-modal knowledge base (MKB), while the other ones are deep generative models adapted from recent state-of-the-art (SOTA) methods. According to our extensive experiments, the key challenges are compositional generalization and effective use of information from both modalities.

Konstantin Dobler, Gerard de Melo

We investigate continued pretraining of LLMs for language adaptation on a tight academic budget: a setting in which only a few GPUs can be used in parallel, for a heavily constrained duration. We focus on adapting Mistral-7B to German or Arabic and evaluate several techniques to improve efficiency and effectiveness in this setting. Our German models adapted on this tight compute budget underperform compared to the base Mistral-7B, while our Arabic models outperform several baselines, showing that for sufficiently well-represented languages, continued pretraining for specialization is not always helpful. Our main findings focus on training precision and tokenizer swapping. Our results show that pure bfloat16 training is a viable alternative to mixed-precision training, while being much faster when only using a few GPUs. Swapping the tokenizer for a specialized one yields more efficient tokenization and is competitive with the original tokenizer, which already contains some German tokens, but did not significantly increase performance for German. Code and model weights are available at on GitHub.

Zengguang Hao, Jie Zhang, Binxia Xu et al.

Intent detection and identification from multi-turn dialogue has become a widely explored technique in conversational agents, for example, voice assistants and intelligent customer services. The conventional approaches typically cast the intent mining process as a classification task. Although neural classifiers have proven adept at such classification tasks, the issue of neural network models often impedes their practical deployment in real-world settings. We present a novel graph-based multi-turn dialogue system called , which identifies a user's intent by identifying intent elements and a standard query from a dynamically constructed and extensible intent graph using reinforcement learning. In addition, we provide visualization components to monitor the immediate reasoning path for each turn of a dialogue, which greatly facilitates further improvement of the system.

Lucas Liebe, Franz Sauerwald, Sylwester Sawicki et al.

Estimating the speed of vehicles using traffic cameras is a crucial task for traffic surveillance and management, enabling more optimal traffic flow, improved road safety, and lower environmental impact. Transportation-dependent systems, such as for navigation and logistics, have great potential to benefit from reliable speed estimation. While there is prior research in this area reporting competitive accuracy levels, their solutions lack reproducibility and robustness across different datasets. To address this, we provide a novel framework for automatic real-time vehicle speed calculation, which copes with more diverse data from publicly available traffic cameras to achieve greater robustness. Our model employs novel techniques to estimate the length of road segments via depth map prediction. Additionally, our framework is capable of handling realistic conditions such as camera movements and different video stream inputs automatically. We compare our model to three well-known models in the field using their benchmark datasets. While our model does not set a new state of the art regarding prediction performance, the results are competitive on realistic CCTV videos. At the same time, our end-to-end pipeline offers more consistent results, an easier implementation, and better compatibility. Its modular structure facilitates reproducibility and future improvements.

Xin Yi, Shunfan Zheng, Linlin Wang et al.

The emergence of finetuning-as-a-service has revealed a new vulnerability in large language models (LLMs). A mere handful of malicious data uploaded by users can subtly manipulate the finetuning process, resulting in an alignment-broken model. Existing methods to counteract fine-tuning attacks typically require substantial computational resources. Even with parameter-efficient techniques like LoRA, gradient updates remain essential. To address these challenges, we propose \textbf{N}euron-\textbf{L}evel \textbf{S}afety \textbf{R}ealignment (\textbf{NLSR}), a training-free framework that restores the safety of LLMs based on the similarity difference of safety-critical neurons before and after fine-tuning. The core of our framework is first to construct a safety reference model from an initially aligned model to amplify safety-related features in neurons. We then utilize this reference model to identify safety-critical neurons, which we prepare as patches. Finally, we selectively restore only those neurons that exhibit significant similarity differences by transplanting these prepared patches, thereby minimally altering the fine-tuned model. Extensive experiments demonstrate significant safety enhancements in fine-tuned models across multiple downstream tasks, while greatly maintaining task-level accuracy. Our findings suggest regions of some safety-critical neurons show noticeable differences after fine-tuning, which can be effectively corrected by transplanting neurons from the reference model without requiring additional training. The code will be available at \url{https://github.com/xinykou/NLSR}

Maximilian Schall, Tamara Czinczoll, Gerard de Melo

Writing commit messages is a tedious daily task for many software developers, and often remains neglected. Automating this task has the potential to save time while ensuring that messages are informative. A high-quality dataset and an objective benchmark are vital preconditions for solid research and evaluation towards this goal. We show that existing datasets exhibit various problems, such as the quality of the commit selection, small sample sizes, duplicates, privacy issues, and missing licenses for redistribution. This can lead to unusable models and skewed evaluations, where inferior models achieve higher evaluation scores due to biases in the data. We compile a new large-scale dataset, CommitBench, adopting best practices for dataset creation. We sample commits from diverse projects with licenses that permit redistribution and apply our filtering and dataset enhancements to improve the quality of generated commit messages. We use CommitBench to compare existing models and show that other approaches are outperformed by a Transformer model pretrained on source code. We hope to accelerate future research by publishing the source code( https://github.com/Maxscha/commitbench ).

Antonio Rueda-Toicen, Abigail Allen Martin, Daniil Morozov et al.

Jaguar re-identification (re-ID) from citizen-science imagery can look strong on standard retrieval metrics while still relying on the wrong evidence, such as background context or silhouette shape, instead of the coat pattern that defines identity. We introduce a diagnostic framework for wildlife re-ID with two axes: a leakage-controlled context ratio, background/foreground, computed from inpainted background-only versus foreground-only images, and a laterality diagnostic based on cross-flank retrieval and mirror self-similarity. To make these diagnostics measurable, we curate a Pantanal jaguar benchmark with per-pixel segmentation masks and an identity-balanced evaluation protocol. We then use representative mitigation families, ArcFace fine-tuning, anti-symmetry regularization, and Lorentz hyperbolic embeddings, as case studies under the same evaluation lens. The goal is not only to ask which model ranks best, but also what visual evidence it uses to do so.

Xiechi Zhang, Shunfan Zheng, Linlin Wang et al.

As multimodal large language models (MLLMs) gain prominence in the medical field, the need for precise evaluation methods to assess their effectiveness has become critical. While benchmarks provide a reliable means to evaluate the capabilities of MLLMs, traditional metrics like ROUGE and BLEU employed for open domain evaluation only focus on token overlap and may not align with human judgment. Although human evaluation is more reliable, it is labor-intensive, costly, and not scalable. LLM-based evaluation methods have proven promising, but to date, there is still an urgent need for open-source multimodal LLM-based evaluators in the medical field. To address this issue, we introduce ACE-$M^3$, an open-sourced \textbf{A}utomatic \textbf{C}apability \textbf{E}valuator for \textbf{M}ultimodal \textbf{M}edical \textbf{M}odels specifically designed to assess the question answering abilities of medical MLLMs. It first utilizes a branch-merge architecture to provide both detailed analysis and a concise final score based on standard medical evaluation criteria. Subsequently, a reward token-based direct preference optimization (RTDPO) strategy is incorporated to save training time without compromising performance of our model. Extensive experiments have demonstrated the effectiveness of our ACE-$M^3$ model\footnote{\url{https://huggingface.co/collections/AIUSRTMP/ace-m3-67593297ff391b93e3e5d068}} in evaluating the capabilities of medical MLLMs.

Margarita Bugueño, Hazem Abou Hamdan, Gerard de Melo

Heterogeneous graph neural networks have recently gained attention for long document summarization, modeling the extraction as a node classification task. Although effective, these models often require external tools or additional machine learning models to define graph components, producing highly complex and less intuitive structures. We present GraphLSS, a heterogeneous graph construction for long document extractive summarization, incorporating Lexical, Structural, and Semantic features. It defines two levels of information (words and sentences) and four types of edges (sentence semantic similarity, sentence occurrence order, word in sentence, and word semantic similarity) without any need for auxiliary learning models. Experiments on two benchmark datasets show that GraphLSS is competitive with top-performing graph-based methods, outperforming recent non-graph models. We release our code on GitHub.

Yindong Wang, Martin Preiß, Margarita Bugueño et al.

Large Language Models (LLMs) frequently confabulate scientific facts, severely undermining their trustworthiness. Addressing this challenge requires benchmarks that go beyond binary factuality and enable fine-grained evaluation. We introduce ReFACT (Reddit False And Correct Texts), a benchmark of 1,001 expert-annotated question-answer pairs spanning diverse scientific domains for the detection of scientific confabulation. Each instance includes both a scientifically correct answer and a non-factual counterpart annotated with precise error spans and error types. ReFACT enables multi-stage evaluation: (1) confabulation detection, (2) fine-grained error localization, and (3) correction. We benchmark 9 state-of-the-art LLMs, revealing limited performance (about 50 percent accuracy). Even top models such as GPT-4o fail to distinguish factual from confabulated scientific answers, raising concerns about the reliability of LLM-as-judge evaluation paradigms. Our findings highlight the need for fine-grained, human-validated benchmarks to detect and correct scientific confabulation in domain-specific contexts. The dataset is available at: https://github.com/ddz5431/ReFACT

Weixing Wang, Zifeng Ding, Jindong Gu et al.

Large Vision-Language Models (LVLMs) with discrete image tokenizers unify multimodal representations by encoding visual inputs into a finite set of tokens. Despite their effectiveness, we find that these models still hallucinate non-existent objects. We hypothesize that this may be due to visual priors induced during training: When certain image tokens frequently co-occur in the same spatial regions and represent shared objects, they become strongly associated with the verbalizations of those objects. As a result, the model may hallucinate by evoking visually absent tokens that often co-occur with present ones. To test this assumption, we construct a co-occurrence graph of image tokens using a segmentation dataset and employ a Graph Neural Network (GNN) with contrastive learning followed by a clustering method to group tokens that frequently co-occur in similar visual contexts. We find that hallucinations predominantly correspond to clusters whose tokens dominate the input, and more specifically, that the visually absent tokens in those clusters show much higher correlation with hallucinated objects compared to tokens present in the image. Based on this observation, we propose a hallucination mitigation method that suppresses the influence of visually absent tokens by modifying latent image embeddings during generation. Experiments show our method reduces hallucinations while preserving expressivity. Code is available at https://github.com/weixingW/CGC-VTD/tree/main

Xiechi Zhang, Zetian Ouyang, Linlin Wang et al.

With the proliferation of large language models (LLMs) in the medical domain, there is increasing demand for improved evaluation techniques to assess their capabilities. However, traditional metrics like F1 and ROUGE, which rely on token overlaps to measure quality, significantly overlook the importance of medical terminology. While human evaluation tends to be more reliable, it can be very costly and may as well suffer from inaccuracies due to limits in human expertise and motivation. Although there are some evaluation methods based on LLMs, their usability in the medical field is limited due to their proprietary nature or lack of expertise. To tackle these challenges, we present AutoMedEval, an open-sourced automatic evaluation model with 13B parameters specifically engineered to measure the question-answering proficiency of medical LLMs. The overarching objective of AutoMedEval is to assess the quality of responses produced by diverse models, aspiring to significantly reduce the dependence on human evaluation. Specifically, we propose a hierarchical training method involving curriculum instruction tuning and an iterative knowledge introspection mechanism, enabling AutoMedEval to acquire professional medical assessment capabilities with limited instructional data. Human evaluations indicate that AutoMedEval surpasses other baselines in terms of correlation with human judgments.

Sedigheh Eslami, Gerard de Melo, Christoph Meinel

Contrastive Language--Image Pre-training (CLIP) has shown remarkable success in learning with cross-modal supervision from extensive amounts of image--text pairs collected online. Thus far, the effectiveness of CLIP has been investigated primarily in general-domain multimodal problems. This work evaluates the effectiveness of CLIP for the task of Medical Visual Question Answering (MedVQA). To this end, we present PubMedCLIP, a fine-tuned version of CLIP for the medical domain based on PubMed articles. Our experiments are conducted on two MedVQA benchmark datasets and investigate two MedVQA methods, MEVF (Mixture of Enhanced Visual Features) and QCR (Question answering via Conditional Reasoning). For each of these, we assess the merits of visual representation learning using PubMedCLIP, the original CLIP, and state-of-the-art MAML (Model-Agnostic Meta-Learning) networks pre-trained only on visual data. We open source the code for our MedVQA pipeline and pre-training PubMedCLIP. CLIP and PubMedCLIP achieve improvements in comparison to MAML's visual encoder. PubMedCLIP achieves the best results with gains in the overall accuracy of up to 3%. Individual examples illustrate the strengths of PubMedCLIP in comparison to the previously widely used MAML networks. Visual representation learning with language supervision in PubMedCLIP leads to noticeable improvements for MedVQA. Our experiments reveal distributional differences in the two MedVQA benchmark datasets that have not been imparted in previous work and cause different back-end visual encoders in PubMedCLIP to exhibit different behavior on these datasets. Moreover, we witness fundamental performance differences of VQA in general versus medical domains.

Honglu Zhou, Advith Chegu, Samuel S. Sohn et al.

Digraph Representation Learning (DRL) aims to learn representations for directed homogeneous graphs (digraphs). Prior work in DRL is largely constrained (e.g., limited to directed acyclic graphs), or has poor generalizability across tasks (e.g., evaluated solely on one task). Most Graph Neural Networks (GNNs) exhibit poor performance on digraphs due to the neglect of modeling neighborhoods and preserving asymmetry. In this paper, we address these notable challenges by leveraging hyperbolic collaborative learning from multi-ordered and partitioned neighborhoods, and regularizers inspired by socio-psychological factors. Our resulting formalism, Digraph Hyperbolic Networks (D-HYPR) - albeit conceptually simple - generalizes to digraphs where cycles and non-transitive relations are common, and is applicable to multiple downstream tasks including node classification, link presence prediction, and link property prediction. In order to assess the effectiveness of D-HYPR, extensive evaluations were performed across 8 real-world digraph datasets involving 21 prior techniques. D-HYPR statistically significantly outperforms the current state of the art. We release our code at https://github.com/hongluzhou/dhypr

Kaustubh D. Dhole, Varun Gangal, Sebastian Gehrmann et al.

Data augmentation is an important component in the robustness evaluation of models in natural language processing (NLP) and in enhancing the diversity of the data they are trained on. In this paper, we present NL-Augmenter, a new participatory Python-based natural language augmentation framework which supports the creation of both transformations (modifications to the data) and filters (data splits according to specific features). We describe the framework and an initial set of 117 transformations and 23 filters for a variety of natural language tasks. We demonstrate the efficacy of NL-Augmenter by using several of its transformations to analyze the robustness of popular natural language models. The infrastructure, datacards and robustness analysis results are available publicly on the NL-Augmenter repository (https://github.com/GEM-benchmark/NL-Augmenter).

Lei Shi, Kai Shuang, Shijie Geng et al.

Several multi-modality representation learning approaches such as LXMERT and ViLBERT have been proposed recently. Such approaches can achieve superior performance due to the high-level semantic information captured during large-scale multimodal pretraining. However, as ViLBERT and LXMERT adopt visual region regression and classification loss, they often suffer from domain gap and noisy label problems, based on the visual features having been pretrained on the Visual Genome dataset. To overcome these issues, we propose unbiased Contrastive Visual-Linguistic Pretraining (CVLP), which constructs a visual self-supervised loss built upon contrastive learning. We evaluate CVLP on several down-stream tasks, including VQA, GQA and NLVR2 to validate the superiority of contrastive learning on multi-modality representation learning. Our code is available at: https://github.com/ArcherYunDong/CVLP-.

Maximilian Schall, Felix Leonard Knöfel, Noah Elias König et al.

Monitoring critically endangered western lowland gorillas is currently hampered by the immense manual effort required to re-identify individuals from vast archives of camera trap footage. The primary obstacle to automating this process has been the lack of large-scale, "in-the-wild" video datasets suitable for training robust deep learning models. To address this gap, we introduce a comprehensive benchmark with three novel datasets: Gorilla-SPAC-Wild, the largest video dataset for wild primate re-identification to date; Gorilla-Berlin-Zoo, for assessing cross-domain re-identification generalization; and Gorilla-SPAC-MoT, for evaluating multi-object tracking in camera trap footage. Building on these datasets, we present GorillaWatch, an end-to-end pipeline integrating detection, tracking, and re-identification. To exploit temporal information, we introduce a multi-frame self-supervised pretraining strategy that leverages consistency in tracklets to learn domain-specific features without manual labels. To ensure scientific validity, a differentiable adaptation of AttnLRP verifies that our model relies on discriminative biometric traits rather than background correlations. Extensive benchmarking subsequently demonstrates that aggregating features from large-scale image backbones outperforms specialized video architectures. Finally, we address unsupervised population counting by integrating spatiotemporal constraints into standard clustering to mitigate over-segmentation. We publicly release all code and datasets to facilitate scalable, non-invasive monitoring of endangered species

Yue Li, Xin Yi, Dongsheng Shi et al.

Watermarking has emerged as a pivotal solution for content traceability and intellectual property protection in Large Vision-Language Models (LVLMs). However, vision-agnostic watermarks may introduce visually irrelevant tokens and disrupt visual grounding by enforcing indiscriminate pseudo-random biases. Additionally, current vision-specific watermarks rely on a static, one-time estimation of vision critical weights and ignore the weight distribution density when determining the proportion of protected tokens. This design fails to account for dynamic changes in visual dependence during generation and may introduce low-quality tokens in the long tail. To address these challenges, we propose Attention-Guided Dynamic Watermarking (AGMark), a novel framework that embeds detectable signals while strictly preserving visual fidelity. At each decoding step, AGMark first dynamically identifies semantic-critical evidence based on attention weights for visual relevance, together with context-aware coherence cues, resulting in a more adaptive and well-calibrated evidence-weight distribution. It then determines the proportion of semantic-critical tokens by jointly considering uncertainty awareness (token entropy) and evidence calibration (weight density), thereby enabling adaptive vocabulary partitioning to avoid irrelevant tokens. Empirical results confirm that AGMark outperforms conventional methods, observably improving generation quality and yielding particularly strong gains in visual semantic fidelity in the later stages of generation. The framework maintains highly competitive detection accuracy (at least 99.36\% AUC) and robust attack resilience (at least 88.61\% AUC) without sacrificing inference efficiency, effectively establishing a new standard for reliability-preserving multi-modal watermarking.

Laura Manduchi, Clara Meister, Kushagra Pandey et al.

The field of deep generative modeling has grown rapidly in the last few years. With the availability of massive amounts of training data coupled with advances in scalable unsupervised learning paradigms, recent large-scale generative models show tremendous promise in synthesizing high-resolution images and text, as well as structured data such as videos and molecules. However, we argue that current large-scale generative AI models exhibit several fundamental shortcomings that hinder their widespread adoption across domains. In this work, our objective is to identify these issues and highlight key unresolved challenges in modern generative AI paradigms that should be addressed to further enhance their capabilities, versatility, and reliability. By identifying these challenges, we aim to provide researchers with insights for exploring fruitful research directions, thus fostering the development of more robust and accessible generative AI solutions.

Yan Cai, Linlin Wang, Ye Wang et al.

The emergence of various medical large language models (LLMs) in the medical domain has highlighted the need for unified evaluation standards, as manual evaluation of LLMs proves to be time-consuming and labor-intensive. To address this issue, we introduce MedBench, a comprehensive benchmark for the Chinese medical domain, comprising 40,041 questions sourced from authentic examination exercises and medical reports of diverse branches of medicine. In particular, this benchmark is composed of four key components: the Chinese Medical Licensing Examination, the Resident Standardization Training Examination, the Doctor In-Charge Qualification Examination, and real-world clinic cases encompassing examinations, diagnoses, and treatments. MedBench replicates the educational progression and clinical practice experiences of doctors in Mainland China, thereby establishing itself as a credible benchmark for assessing the mastery of knowledge and reasoning abilities in medical language learning models. We perform extensive experiments and conduct an in-depth analysis from diverse perspectives, which culminate in the following findings: (1) Chinese medical LLMs underperform on this benchmark, highlighting the need for significant advances in clinical knowledge and diagnostic precision. (2) Several general-domain LLMs surprisingly possess considerable medical knowledge. These findings elucidate both the capabilities and limitations of LLMs within the context of MedBench, with the ultimate goal of aiding the medical research community.

Roi Cohen, Konstantin Dobler, Eden Biran et al.

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known as hallucinations, causing them to emit unwanted and factually incorrect text. In this work, we propose a novel calibration method that can be used to combat hallucinations. We add a special [IDK] ("I don't know") token to the model's vocabulary and introduce an objective function that shifts probability mass to the [IDK] token for incorrect predictions. This approach allows the model to express uncertainty in its output explicitly. We evaluate our proposed method across multiple model architectures and factual downstream tasks. We find that models trained with our method are able to express uncertainty in places where they would previously make mistakes while suffering only a small loss of encoded knowledge. We further perform extensive ablation studies of multiple variations of our approach and provide a detailed analysis of the precision-recall tradeoff of our method.

Tolga Buz, Benjamin Frost, Nikola Genchev et al.

Recent Large Language Models (LLMs) have shown the ability to generate content that is difficult or impossible to distinguish from human writing. We investigate the ability of differently-sized LLMs to replicate human writing style in short, creative texts in the domain of Showerthoughts, thoughts that may occur during mundane activities. We compare GPT-2 and GPT-Neo fine-tuned on Reddit data as well as GPT-3.5 invoked in a zero-shot manner, against human-authored texts. We measure human preference on the texts across the specific dimensions that account for the quality of creative, witty texts. Additionally, we compare the ability of humans versus fine-tuned RoBERTa classifiers to detect AI-generated texts. We conclude that human evaluators rate the generated texts slightly worse on average regarding their creative quality, but they are unable to reliably distinguish between human-written and AI-generated texts. We further provide a dataset for creative, witty text generation based on Reddit Showerthoughts posts.

Tamara Czinczoll, Christoph Hönes, Maximilian Schall et al.

While (large) language models have significantly improved over the last years, they still struggle to sensibly process long sequences found, e.g., in books, due to the quadratic scaling of the underlying attention mechanism. To address this, we propose NextLevelBERT, a Masked Language Model operating not on tokens, but on higher-level semantic representations in the form of text embeddings. We pretrain NextLevelBERT to predict the vector representation of entire masked text chunks and evaluate the effectiveness of the resulting document vectors on three types of tasks: 1) Semantic Textual Similarity via zero-shot document embeddings, 2) Long document classification, 3) Multiple-choice question answering. We find that next-level Masked Language Modeling is an effective technique to tackle long-document use cases and can outperfor much larger embedding models as long as the required level of detail of semantic information is not too fine. Our models and code are publicly available online.

Konstantin Dobler, Desmond Elliott, Gerard de Melo

Current language models rely on static vocabularies determined at pretraining time, which can lead to decreased performance and increased computational cost for domains underrepresented in the original vocabulary. New tokens can be added to solve this problem, when coupled with a good initialization for their new embeddings. However, existing embedding initialization methods require expensive further training or pretraining of additional modules. In this paper, we propose Token Distillation and show that by distilling representations obtained using the original tokenization, we can quickly learn high-quality input embeddings for new tokens. Experimental results with a wide range of open-weight models show that Token Distillation outperforms even strong baselines.

Shunfan Zheng, Xiechi Zhang, Gerard de Melo et al.

In the rapidly evolving landscape of large language models (LLMs) for medical applications, ensuring the reliability and accuracy of these models in clinical settings is paramount. Existing benchmarks often focus on fixed-format tasks like multiple-choice QA, which fail to capture the complexity of real-world clinical diagnostics. Moreover, traditional evaluation metrics and LLM-based evaluators struggle with misalignment, often providing oversimplified assessments that do not adequately reflect human judgment. To address these challenges, we introduce HDCEval, a Hierarchical Divide-and-Conquer Evaluation framework tailored for fine-grained alignment in medical evaluation. HDCEval is built on a set of fine-grained medical evaluation guidelines developed in collaboration with professional doctors, encompassing Patient Question Relevance, Medical Knowledge Correctness, and Expression. The framework decomposes complex evaluation tasks into specialized subtasks, each evaluated by expert models trained through Attribute-Driven Token Optimization (ADTO) on a meticulously curated preference dataset. This hierarchical approach ensures that each aspect of the evaluation is handled with expert precision, leading to a significant improvement in alignment with human evaluators.

Yue Li, Xin Yi, Dongsheng Shi et al.

Fingerprinting is critical for maintaining traceability and protecting the intellectual property (IP) of developers, as LLMs deployed in web applications are susceptible to unauthorized redistribution and misuse via fine-tuning or black-box deployment. However, current backdoor-based fingerprinting methods face a fundamental trade-off: fingerprints embedded as garbled text are easily detected and filtered, whereas those crafted as coherent natural language are prone to being triggered unintentionally. To overcome these limitations, we propose RFEdit, a knowledge-editing framework that embeds a rule-based multilingual natural language fingerprint (MNLF) by modifying a sparse subset of model weights. This approach enables efficient and robust fingerprint injection with minimal impact on unrelated knowledge in LLMs. Our RFEdit framework is further safeguarded by Fingerprint Subspace-aware Fine-Tuning (FSFT), which mitigates fingerprint degradation during legitimate fine-tuning by restricting parameter updates to the fingerprint subspace. This approach preserves fingerprint integrity while enhancing downstream task performance of LLMs. These advances establish a comprehensive pipeline from fingerprint injection to defense, achieving high detection effectiveness, robustness against adversarial manipulations, harmlessness to model utility, and persistence under fine-tuning. Extensive experiments demonstrate that RFEdit maintains robustness under quantization and pruning. Additionally, fingerprint effectiveness is generally improved by more than 10\% when combined with FSFT for math and alpaca downstream tasks.

Lihu Chen, Gerard de Melo, Fabian M. Suchanek et al.

It is important for Large Language Models (LLMs) to be aware of the boundary of their knowledge, distinguishing queries they can confidently answer from those that lie beyond their capabilities. Such awareness enables models to perform adaptive inference, such as invoking retrieval-augmented generation (RAG), engaging in slow and deep thinking, or abstaining from answering when appropriate. These mechanisms are key to developing efficient and trustworthy AI. In this work, we propose a method to detect knowledge boundaries via Query-Level Uncertainty, which estimates if a model is capable of answering a given query before generating any tokens, thus avoiding the generation cost. To this end, we propose a novel, training-free method called Internal Confidence, which leverages self-evaluations across layers and tokens to provide a reliable signal of uncertainty. Empirical studies on both factual question answering and mathematical reasoning tasks demonstrate that our Internal Confidence outperforms several baselines in quality of confidence while being computationally cheaper. Furthermore, we demonstrate its benefits in adaptive inference settings, showing that for RAG and model cascading it reduces inference costs while preserving overall performance.

Roi Cohen, Omri Fahn, Gerard de Melo

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known as hallucinations, causing them to emit unwanted and factually incorrect text. In this work, we study how well LLMs capture uncertainty, without explicitly being trained for that. We show that, if considering uncertainty as a linear concept in the model's latent space, it might indeed be captured, even after only pretraining. We further show that, though unintuitive, LLMs appear to capture several different types of uncertainty, each of which can be useful to predict the correctness for a specific task or benchmark. Furthermore, we provide in-depth results such as demonstrating a correlation between our correction prediction and the model's ability to abstain from misinformation using words, and the lack of impact of model scaling for capturing uncertainty. Finally, we claim that unifying the uncertainty types as a single one using instruction-tuning or [IDK]-token tuning is helpful for the model in terms of correctness prediction.

Harshil Darji, Martin Heckelmann, Christina Kratsch et al.

The availability of structured legal data is important for advancing Natural Language Processing (NLP) techniques for the German legal system. One of the most widely used datasets, Open Legal Data, provides a large-scale collection of German court decisions. While the metadata in this raw dataset is consistently structured, the decision texts themselves are inconsistently formatted and often lack clearly marked sections. Reliable separation of these sections is important not only for rhetorical role classification but also for downstream tasks such as retrieval and citation analysis. In this work, we introduce a cleaned and sectioned dataset of 251,038 German court decisions derived from the official Open Legal Data dataset. We systematically separated three important sections in German court decisions, namely Tenor (operative part of the decision), Tatbestand (facts of the case), and Entscheidungsgründe (judicial reasoning), which are often inconsistently represented in the original dataset. To ensure the reliability of our extraction process, we used Cochran's formula with a 95% confidence level and a 5% margin of error to draw a statistically representative random sample of 384 cases, and manually verified that all three sections were correctly identified. We also extracted the Rechtsmittelbelehrung (appeal notice) as a separate field, since it is a procedural instruction and not part of the decision itself. The resulting corpus is publicly available in the JSONL format, making it an accessible resource for further research on the German legal system.

Liu Liu, Alexandra Kudaeva, Marco Cipriano et al.

Understanding group-level social interactions in public spaces is crucial for urban planning, informing the design of socially vibrant and inclusive environments. Detecting such interactions from images involves interpreting subtle visual cues such as relations, proximity, and co-movement - semantically complex signals that go beyond traditional object detection. To address this challenge, we introduce a social group region detection task, which requires inferring and spatially grounding visual regions defined by abstract interpersonal relations. We propose MINGLE (Modeling INterpersonal Group-Level Engagement), a modular three-stage pipeline that integrates: (1) off-the-shelf human detection and depth estimation, (2) VLM-based reasoning to classify pairwise social affiliation, and (3) a lightweight spatial aggregation algorithm to localize socially connected groups. To support this task and encourage future research, we present a new dataset of 100K urban street-view images annotated with bounding boxes and labels for both individuals and socially interacting groups. The annotations combine human-created labels and outputs from the MINGLE pipeline, ensuring semantic richness and broad coverage of real-world scenarios.

Margarita Bugueño, Gerard de Melo

In document classification, graph-based models effectively capture document structure, overcoming sequence length limitations and enhancing contextual understanding. However, most existing graph document representations rely on heuristics, domain-specific rules, or expert knowledge. Unlike previous approaches, we propose a method to learn data-driven graph structures, eliminating the need for manual design and reducing domain dependence. Our approach constructs homogeneous weighted graphs with sentences as nodes, while edges are learned via a self-attention model that identifies dependencies between sentence pairs. A statistical filtering strategy aims to retain only strongly correlated sentences, improving graph quality while reducing the graph size. Experiments on three document classification datasets demonstrate that learned graphs consistently outperform heuristic-based graphs, achieving higher accuracy and $F_1$ score. Furthermore, our study demonstrates the effectiveness of the statistical filtering in improving classification robustness. These results highlight the potential of automatic graph generation over traditional heuristic approaches and open new directions for broader applications in NLP.

Yue Li, Xin Yi, Dongsheng Shi et al.

With the increasing size of Large Vision-Language Models (LVLMs), network pruning techniques aimed at compressing models for deployment in resource-constrained environments have garnered significant attention. However, we observe that pruning often leads to a degradation in safety performance. To address this issue, we present a novel and lightweight approach, termed Hierarchical Safety Realignment (HSR). HSR operates by first quantifying the contribution of each attention head to safety, identifying the most critical ones, and then selectively restoring neurons directly within these attention heads that play a pivotal role in maintaining safety. This process hierarchically realigns the safety of pruned LVLMs, progressing from the attention head level to the neuron level. We validate HSR across various models and pruning strategies, consistently achieving notable improvements in safety performance. To our knowledge, this is the first work explicitly focused on restoring safety in LVLMs post-pruning.

Leo Kohlenberg, Leonard Horns, Frederic Sadrieh et al.

Annotating large datasets can be challenging. However, crowd-sourcing is often expensive and can lack quality, especially for non-trivial tasks. We propose a method of using LLMs as few-shot learners for annotating data in a complex natural language task where we learn a standalone model to predict usage options for products from customer reviews. We also propose a new evaluation metric for this scenario, HAMS4, that can be used to compare a set of strings with multiple reference sets. Learning a custom model offers individual control over energy efficiency and privacy measures compared to using the LLM directly for the sequence-to-sequence task. We compare this data annotation approach with other traditional methods and demonstrate how LLMs can enable considerable cost savings. We find that the quality of the resulting data exceeds the level attained by third-party vendor services and that GPT-4-generated labels even reach the level of domain experts. We make the code and generated labels publicly available.

Sedigheh Eslami, Gerard de Melo

Contrastive Language--Image Pre-training (CLIP) has manifested remarkable improvements in zero-shot classification and cross-modal vision-language tasks. Yet, from a geometrical point of view, the CLIP embedding space has been found to have a pronounced modality gap. This gap renders the embedding space overly sparse and disconnected, with different modalities being densely distributed in distinct subregions of the hypersphere. In this work, we aim at answering three main questions: 1. Does sharing the parameter space between the multi-modal encoders reduce the modality gap? 2. Can the gap be mitigated by pushing apart the uni-modal embeddings via intra-modality separation? 3. How do these gap reduction approaches affect the downstream performance? We design AlignCLIP, in order to answer these questions and through extensive experiments, we show that AlignCLIP achieves noticeable enhancements in the cross-modal alignment of the embeddings, and thereby, reduces the modality gap, while improving the performance across several zero-shot and fine-tuning downstream evaluations.

Maryam Hosseini, Marco Cipriano, Sedigheh Eslami et al.

Existing Open Vocabulary Detection (OVD) models exhibit a number of challenges. They often struggle with semantic consistency across diverse inputs, and are often sensitive to slight variations in input phrasing, leading to inconsistent performance. The calibration of their predictive confidence, especially in complex multi-label scenarios, remains suboptimal, frequently resulting in overconfident predictions that do not accurately reflect their context understanding. To understand these limitations, multi-label detection benchmarks are needed. A particularly challenging domain for such benchmarking is social activities. Due to the lack of multi-label benchmarks for social interactions, in this work we present ELSA: Evaluating Localization of Social Activities. ELSA draws on theoretical frameworks in urban sociology and design and uses in-the-wild street-level imagery, where the size of groups and the types of activities vary significantly. ELSA includes more than 900 manually annotated images with more than 4,300 multi-labeled bounding boxes for individual and group activities. We introduce a novel confidence score computation method NLSE and a novel Dynamic Box Aggregation (DBA) algorithm to assess semantic consistency in overlapping predictions. We report our results on the widely-used SOTA models Grounding DINO, Detic, OWL, and MDETR. Our evaluation protocol considers semantic stability and localization accuracy and further exposes the limitations of existing approaches.

Margarita Bugueño, Gerard de Melo

Given the success of Graph Neural Networks (GNNs) for structure-aware machine learning, many studies have explored their use for text classification, but mostly in specific domains with limited data characteristics. Moreover, some strategies prior to GNNs relied on graph mining and classical machine learning, making it difficult to assess their effectiveness in modern settings. This work extensively investigates graph representation methods for text classification, identifying practical implications and open challenges. We compare different graph construction schemes using a variety of GNN architectures and setups across five datasets, encompassing short and long documents as well as unbalanced scenarios in diverse domains. Two Transformer-based large language models are also included to complement the study. The results show that i) although the effectiveness of graphs depends on the textual input features and domain, simple graph constructions perform better the longer the documents are, ii) graph representations are especially beneficial for longer documents, outperforming Transformer-based models, iii) graph methods are particularly efficient at solving the task.

Konstantin Dobler, Gerard de Melo

Using model weights pretrained on a high-resource language as a warm start can reduce the need for data and compute to obtain high-quality language models for other, especially low-resource, languages. However, if we want to use a new tokenizer specialized for the target language, we cannot transfer the source model's embedding matrix. In this paper, we propose FOCUS - Fast Overlapping Token Combinations Using Sparsemax, a novel embedding initialization method that initializes the embedding matrix effectively for a new tokenizer based on information in the source model's embedding matrix. FOCUS represents newly added tokens as combinations of tokens in the overlap of the source and target vocabularies. The overlapping tokens are selected based on semantic similarity in an auxiliary static token embedding space. We focus our study on using the multilingual XLM-R as a source model and empirically show that FOCUS outperforms random initialization and previous work in language modeling and on a range of downstream tasks (NLI, QA, and NER).

Konstantin Dobler, Florian Hübscher, Jan Westphal et al.

Creating meaningful art is often viewed as a uniquely human endeavor. A human artist needs a combination of unique skills, understanding, and genuine intention to create artworks that evoke deep feelings and emotions. In this paper, we introduce a multi-conditional Generative Adversarial Network (GAN) approach trained on large amounts of human paintings to synthesize realistic-looking paintings that emulate human art. Our approach is based on the StyleGAN neural network architecture, but incorporates a custom multi-conditional control mechanism that provides fine-granular control over characteristics of the generated paintings, e.g., with regard to the perceived emotion evoked in a spectator. For better control, we introduce the conditional truncation trick, which adapts the standard truncation trick for the conditional setting and diverse datasets. Finally, we develop a diverse set of evaluation techniques tailored to multi-conditional generation.

Lei Shi, Kai Shuang, Shijie Geng et al.

Inspired by the success of BERT, several multimodal representation learning approaches have been proposed that jointly represent image and text. These approaches achieve superior performance by capturing high-level semantic information from large-scale multimodal pretraining. In particular, LXMERT and UNITER adopt visual region feature regression and label classification as pretext tasks. However, they tend to suffer from the problems of noisy labels and sparse semantic annotations, based on the visual features having been pretrained on a crowdsourced dataset with limited and inconsistent semantic labeling. To overcome these issues, we propose unbiased Dense Contrastive Visual-Linguistic Pretraining (DCVLP), which replaces the region regression and classification with cross-modality region contrastive learning that requires no annotations. Two data augmentation strategies (Mask Perturbation and Intra-/Inter-Adversarial Perturbation) are developed to improve the quality of negative samples used in contrastive learning. Overall, DCVLP allows cross-modality dense region contrastive learning in a self-supervised setting independent of any object annotations. We compare our method against prior visual-linguistic pretraining frameworks to validate the superiority of dense contrastive learning on multimodal representation learning.

Xiang Hu, Haitao Mi, Zujie Wen et al.

Human language understanding operates at multiple levels of granularity (e.g., words, phrases, and sentences) with increasing levels of abstraction that can be hierarchically combined. However, existing deep models with stacked layers do not explicitly model any sort of hierarchical process. This paper proposes a recursive Transformer model based on differentiable CKY style binary trees to emulate the composition process. We extend the bidirectional language model pre-training objective to this architecture, attempting to predict each word given its left and right abstraction nodes. To scale up our approach, we also introduce an efficient pruned tree induction algorithm to enable encoding in just a linear number of composition steps. Experimental results on language modeling and unsupervised parsing show the effectiveness of our approach.