LLM-based judges have emerged as a scalable alternative to human evaluation and are increasingly used to assess, compare, and improve models. However, the reliability of LLM-based judges themselves is rarely scrutinized. As LLMs become more advanced, their responses grow more sophisticated, requiring stronger judges to evaluate them. Existing benchmarks primarily focus on a judge’s alignment with human preferences, but often fail to account for more challenging tasks where crowdsourced human preference is a poor indicator of factual and logical correctness. To address this, we propose a novel evaluation framework to objectively evaluate LLM-based judges. Based on this framework, we propose JudgeBench, a benchmark for evaluating LLM-based judges on challenging response pairs spanning knowledge, reasoning, math, and coding. JudgeBench leverages a novel pipeline for converting existing difficult datasets into challenging response pairs with preference labels reflecting objective correctness. Our comprehensive evaluation on a collection of prompted judges, fine-tuned judges, multi-agent judges, and reward models shows that JudgeBench poses a significantly greater challenge than previous benchmarks, with many strong models (e.g. GPT-4o) performing just slightly better than random guessing. Overall, JudgeBench offers a reliable platform for assessing increasingly advanced LLM-based judges. Data and code are available at \url{https://github.com/ScalerLab/JudgeBench}.

This paper presents an analytical study of the oversmoothing issue in diffusion-based Graph Neural Networks (GNNs). Generalizing beyond extant approaches grounded in random walk analysis or particle systems, we approach this problem through operator semigroup theory. This theoretical framework allows us to rigorously prove that oversmoothing is intrinsically linked to the ergodicity of the diffusion operator. Relying on semigroup method, we can quantitatively analyze the dynamic of graph diffusion and give a specific mathematical form of the smoothing feature by ergodicity and invariant measure of operator, which improves previous works only show existence of oversmoothing. This finding further poses a general and mild ergodicity-breaking condition, encompassing the various specific solutions previously offered, thereby presenting a more universal and theoretically grounded approach to relieve oversmoothing in diffusion-based GNNs. Additionally, we offer a probabilistic interpretation of our theory, forging a link with prior works and broadening the theoretical horizon. Our experimental results reveal that this ergodicity-breaking term effectively mitigates oversmoothing measured by Dirichlet energy, and simultaneously enhances performance in node classification tasks.

As foundation models become more sophisticated, ensuring their trustworthiness becomes increasingly critical; yet, unlike text and image, the video modality still lacks comprehensive trustworthiness benchmarks. We introduce VMDT (Video-Modal DecodingTrust), the first unified platform for evaluating text-to-video (T2V) and video-to-text (V2T) models across five key trustworthiness dimensions: safety, hallucination, fairness, privacy, and adversarial robustness. Through our extensive evaluation of 7 T2V models and 19 V2T models using VMDT, we uncover several significant insights. For instance, all open-source T2V models evaluated fail to recognize harmful queries and often generate harmful videos, while exhibiting higher levels of unfairness compared to image modality models. In V2T models, unfairness and privacy risks rise with scale, whereas hallucination and adversarial robustness improve---though overall performance remains low. Uniquely, safety shows no correlation with model size, implying that factors other than scale govern current safety levels. Our findings highlight the urgent need for developing more robust and trustworthy video foundation models, and VMDT provides a systematic framework for measuring and tracking progress toward this goal. The code is available at https://sunblaze-ucb.github.io/VMDT-page/.

We present a new method for large language models to solve compositional tasks.Although they have shown strong performance on traditional language understanding tasks, large language models struggle to solve compositional tasks, where the solution depends on solving smaller instances of the same problem.We propose a natural approach to solve compositional tasks recursively.Our method, Re-Tuning, tunes models to break down a problem into subproblems, solve those subproblems, and combine the results.We show that our method significantly improves model performance on three representative compositional tasks: integer addition, dynamic programming, and parity.Compared to state-of-the-art methods that keep intermediate steps towards solving the problems, Re-Tuning achieves significantly higher accuracy and is more GPU memory efficient.

We introduce a new challenge to test the STEM skills of neural models. The problems in the real world often require solutions, combining knowledge from STEM (science, technology, engineering, and math). Unlike existing datasets, our dataset requires the understanding of multimodal vision-language information of STEM. Our dataset features one of the largest and most comprehensive datasets for the challenge. It includes $448$ skills and $1,073,146$ questions spanning all STEM subjects. Compared to existing datasets that often focus on examining expert-level ability, our dataset includes fundamental skills and questions designed based on the K-12 curriculum. We also add state-of-the-art foundation models such as CLIP and GPT-3.5-Turbo to our benchmark. Results show that the recent model advances only help master a very limited number of lower grade-level skills ($2.5$% in the third grade) in our dataset. In fact, these models are still well below (averaging $54.7$%) the performance of elementary students, not to mention near expert-level performance. To understand and increase the performance on our dataset, we teach the models on a training split of our dataset. Even though we observe improved performance, the model performance remains relatively low compared to average elementary students. To solve STEM problems, we will need novel algorithmic innovations from the community.

We introduce a method to improve the zero-shot reasoning abilities of large language models on general language understanding tasks. Specifically, we build an autonomous agent to instruct the reasoning process of large language models. To enable this, our agent only needs to generate a single set of instructions for each task. These instructions turn out to be extremely effective for improving the reasoning process of different large language models across all task instances. We show this approach further unleashes the zero-shot reasoning abilities of large language models to more tasks. We study the performance of our method on a wide set of datasets spanning generation, classification, and reasoning. We show that our method generalizes to most tasks and obtains state-of-the-art zero-shot performance on 20 of the 29 datasets that we evaluate. For instance, our method boosts the performance of state-of-the-art large language models by a large margin, including Vicuna-13b, Llama-2-70b-chat, and GPT-3.5 Turbo. Compared to zero-shot chain of thought, our improvement in reasoning is striking. With our method, Llama-2-70b-chat outperforms zero-shot GPT-3.5 Turbo significantly.

Membership inference attacks (MIAs) aim to infer whether a data point has been used to train a machine learning model. These attacks can be employed to identify potential privacy vulnerabilities and detect unauthorized use of personal data. While MIAs have been traditionally studied for simple classification models, recent advancements in multi-modal pre-training, such as CLIP, have demonstrated remarkable zero-shot performance across a range of computer vision tasks. However, the sheer scale of data and models presents significant computational challenges for performing the attacks.This paper takes a first step towards developing practical MIAs against large-scale multi-modal models. We introduce a simple baseline strategy by thresholding the cosine similarity between text and image features of a target point and propose further enhancing the baseline by aggregating cosine similarity across transformations of the target. We also present a new weakly supervised attack method that leverages ground-truth non-members (e.g., obtained by using the publication date of a target model and the timestamps of the open data) to further enhance the attack. Our evaluation shows that CLIP models are susceptible to our attack strategies, with our simple baseline achieving over 75% membership identification accuracy. Furthermore, our enhanced attacks outperform the baseline across multiple models and datasets, with the weakly supervised attack demonstrating an average-case performance improvement of 17% and being at least 7X more effective at low false-positive rates. These findings highlight the importance of protecting the privacy of multi-modal foundational models, which were previously assumed to be less susceptible to MIAs due to less overfitting. Our code is available at https://github.com/ruoxi-jia-group/CLIP-MIA.

Nowadays, due to the breakthrough in natural language generation (NLG), including machine translation, document summarization, image captioning, etc NLG models have been encapsulated in cloud APIs to serve over half a billion people worldwide and process over one hundred billion word generations per day. Thus, NLG APIs have already become essential profitable services in many commercial companies. Due to the substantial financial and intellectual investments, service providers adopt a pay-as-you-use policy to promote sustainable market growth. However, recent works have shown that cloud platforms suffer from financial losses imposed by model extraction attacks, which aim to imitate the functionality and utility of the victim services, thus violating the intellectual property (IP) of cloud APIs. This work targets at protecting IP of NLG APIs by identifying the attackers who have utilized watermarked responses from the victim NLG APIs. However, most existing watermarking techniques are not directly amenable for IP protection of NLG APIs. To bridge this gap, we first present a novel watermarking method for text generation APIs by conducting lexical modification to the original outputs. Compared with the competitive baselines, our watermark approach achieves better identifiable performance in terms of p-value, with fewer semantic losses. In addition, our watermarks are more understandable and intuitive to humans than the baselines. Finally, the empirical studies show our approach is also applicable to queries from different domains, and is effective on the attacker trained on a mixture of the corpus which includes less than 10% watermarked samples.

The task of completing knowledge triplets has broad downstream applications. Both structural and semantic information plays an important role in knowledge graph completion. Unlike previous approaches that rely on either the structures or semantics of the knowledge graphs, we propose to jointly embed the semantics in the natural language description of the knowledge triplets with their structure information. Our method embeds knowledge graphs for the completion task via fine-tuning pre-trained language models with respect to a probabilistic structured loss, where the forward pass of the language models captures semantics and the loss reconstructs structures. Our extensive experiments on a variety of knowledge graph benchmarks have demonstrated the state-of-the-art performance of our method. We also show that our method can significantly improve the performance in a low-resource regime, thanks to the better use of semantics. The code and datasets are available at https://github.com/pkusjh/LASS.

We introduce a new open information extraction (OIE) benchmark for pre-trained language models (LM).Recent studies have demonstrated that pre-trained LMs, such as BERT and GPT, may store linguistic and relational knowledge.In particular, LMs are able to answer "fill-in-the-blank" questions when given a pre-defined relation category.Instead of focusing on pre-defined relations, we create an OIE benchmark aiming to fully examine the open relational information present in the pretrained LMs.We accomplish this by turning pre-trained LMs into zero-shot OIE systems.Surprisingly, pre-trained LMs are able to obtain competitive performance on both standard OIE datasets (CaRB and Re-OIE2016) and two new large-scale factual OIE datasets (TAC KBP-OIE and Wikidata-OIE) that we establish via distant supervision.For instance, the zeroshot pre-trained LMs outperform the F1 score of the state-of-the-art supervised OIE methods on our factual OIE datasets without needing to use any training sets. 1 Dylan was in born Minnesota

We cast a suite of information extraction tasks into a text-to-triple translation framework.Instead of solving each task relying on taskspecific datasets and models, we formalize the task as a translation between task-specific input text and output triples.By taking the taskspecific input, we enable a task-agnostic translation by leveraging the latent knowledge that a pre-trained language model has about the task.We further demonstrate that a simple pretraining task of predicting which relational information corresponds to which input text is an effective way to produce task-specific outputs.This enables the zero-shot transfer of our framework to downstream tasks.We study the zero-shot performance of this framework on open information extraction (OIE2016, NYT, WEB, PENN), relation classification (FewRel and TACRED), and factual probe (Google-RE and T-REx).The model transfers non-trivially to most tasks and is often competitive with a fully supervised method without the need for any task-specific training.For instance, we significantly outperform the F1 score of the supervised open information extraction without needing to use its training set.

Dependency context-based word embedding jointly learns the representations of word and dependency context, and has been proved effective in aspect term extraction.In this paper, we design the positional dependency-based word embedding (POD) which considers both dependency context and positional context for aspect term extraction.Specifically, the positional context is modeled via relative position encoding.Besides, we enhance the dependency context by integrating more lexical information (e.g., POS tags) along dependency paths.Experiments on SemEval 2014/2015/2016 datasets show that our approach outperforms other embedding methods in aspect term extraction.

We present GluonCV and GluonNLP, the deep learning toolkits for computer vision and natural language processing based on Apache MXNet (incubating). These toolkits provide state-of-the-art pre-trained models, training scripts, and training logs, to facilitate rapid prototyping and promote reproducible research. We also provide modular APIs with flexible building blocks to enable efficient customization. Leveraging the MXNet ecosystem, the deep learning models in GluonCV and GluonNLP can be deployed onto a variety of platforms with different programming languages. The Apache 2.0 license has been adopted by GluonCV and GluonNLP to allow for software distribution, modification, and usage.

Recent work has achieved great success in utilizing global contextual information for semantic segmentation, including increasing the receptive field and aggregating pyramid feature representations. In this paper, we go beyond global context and explore the fine-grained representation using co-occurrent features by introducing Co-occurrent Feature Model, which predicts the distribution of co-occurrent features for a given target. To leverage the semantic context in the co-occurrent features, we build an Aggregated Co-occurrent Feature (ACF) Module by aggregating the probability of the co-occurrent feature with the co-occurrent context. ACF Module learns a fine-grained spatial invariant representation to capture co-occurrent context information across the scene. Our approach significantly improves the segmentation results using FCN and achieves superior performance 54.0% mIoU on Pascal Context, 87.2% mIoU on Pascal VOC 2012 and 44.89% mIoU on ADE20K datasets. The source code and complete system will be publicly available upon publication.

Measuring network similarity is a fundamental data mining problem. The mainstream similarity measures mainly leverage the structural information regarding to the entities in the network without considering the network semantics. In the real world, the heterogeneous information networks (HINs) with rich semantics are ubiquitous. However, the existing network similarity doesn't generalize well in HINs because they fail to capture the HIN semantics. The meta-path has been proposed and demonstrated as a right way to represent semantics in HINs. Therefore, original meta-path based similarities (e.g., PathSim and KnowSim) have been successful in computing the entity proximity in HINs. The intuition is that the more instances of meta-path(s) between entities, the more similar the entities are. Thus the original meta-path similarity only applies to computing the proximity of two neighborhood (connected) entities. In this paper, we propose the distant meta-path similarity that is able to capture HIN semantics between two distant (isolated) entities to provide more meaningful entity proximity. The main idea is that even there is no shared neighborhood entities of (i.e., no meta-path instances connecting) the two entities, but if the more similar neighborhood entities of the entities are, the more similar the two entities should be. We then find out the optimum distant meta-path similarity by exploring the similarity hypothesis space based on different theoretical foundations. We show the state-of-the-art similarity performance of distant meta-path similarity on two text-based HINs and make the datasets public available.

Crowdsourcing has proven to be an effective method for generating labeled data for a range of NLP tasks.However, multiple recent attempts of using crowdsourcing to generate gold-labeled training data for semantic role labeling (SRL) reported only modest results, indicating that SRL is perhaps too difficult a task to be effectively crowdsourced.In this paper, we postulate that while producing SRL annotation does require expert involvement in general, a large subset of SRL labeling tasks is in fact appropriate for the crowd.We present a novel workflow in which we employ a classifier to identify difficult annotation tasks and route each task either to experts or crowd workers according to their difficulties.Our experimental evaluation shows that the proposed approach reduces the workload for experts by over two-thirds, and thus significantly reduces the cost of producing SRL annotation at little loss in quality.

Heterogeneous information networks (HINs) is a general representation of many real world applications. The difference between HIN and traditional homogeneous graphs is that the nodes and edges in HIN are with types. Then in the many applications, we need to consider the types to make the approach more semantically meaningful. For the applications that annotation is expensive, on natural way is to consider semi-supervised learning over HIN. In this paper, we present a semi-supervised learning algorithm constrained by the types of HINs. We first decompose the original HIN into several semantically meaningful sub-graphs based the meta-graphs composed of entity and relation types. Then we perform random walk over the sub-graphs to propagate the labels from labeled data to unlabeled data. After we obtain all the labels propagated by different trials of random walk guided by meta-graphs, we use an ensemble algorithm to vote for the final labeling results. We use two public available datasets, 20-newsgroups and RCV1 datasets to test our algorithm. Experimental results show that our algorithm is better than the traditional semi-supervised learning algorithms for HINs. One particular by-product of this work is that we show that previous random walk approach guided by meta-paths can be non-stationary, which is the major reason we propose a meta-graph guide random walk for semi-supervised learning over HINs.

Active learning is a useful technique for tasks for which unlabeled data is abundant but manual labeling is expensive. One example of such a task is semantic role labeling (SRL), which relies heavily on labels from trained linguistic experts. One challenge in applying active learning algorithms for SRL is that the complete knowledge of the SRL model is often unavailable, against the common assumption that active learning methods are aware of the details of the underlying models. In this paper, we present an active learning framework for black-box SRL models (i.e., models whose details are unknown). In lieu of a query strategy based on model details, we propose a neural query strategy model that embeds both language and semantic information to automatically learn the query strategy from predictions of an SRL model alone. Our experimental results demonstrate the effectiveness of both this new active learning framework and the neural query strategy model.

Text classification is an important problem with many applications. Traditional approaches represent text as a bag-of-words and build classifiers based on this representation. Rather than words, entity phrases, the relations between the entities, as well as the types of the entities and relations carry much more information to represent the texts. This paper presents a novel text as network classification framework, which introduces 1) a structured and typed heterogeneous information networks (HINs) representation of texts, and 2) a meta-path based approach to link texts. We show that with the new representation and links of texts, the structured and typed information of entities and relations can be incorporated into kernels. Particularly, we develop both simple linear kernel and indefinite kernel based on meta-paths in the HIN representation of texts, where we call them HIN-kernels. Using Freebase, a well-known world knowledge base, to construct HIN for texts, our experiments on two benchmark datasets show that the indefinite HIN kernel based on weighted meta-paths outperforms the state-of-the-art methods and other HIN-kernels.

Recent studies have demonstrated the power of modeling real world data as heterogeneous information networks (HINs) consisting of multiple types of entities and relations. Unfortunately, most of such studies (e.g., similarity search) confine discussions on the networks with only a few entity and relationship types, such as DBLP. In the real world, however, the network schema can be rather complex, such as Freebase. In such HINs with rich schema, it is often too much burden to ask users to provide explicit guidance in selecting relations for similarity search. In this paper, we study the problem of relation similarity search in schema-rich HINs. Under our problem setting, users are only asked to provide some simple relation instance examples (e.g., 〈Barack Obama, John Kerry〉 and 〈George W. Bush, Condoleezza Rice〉) as a query, and we automatically detect the latent semantic relation (LSR) implied by the query (e.g., “president vs. secretary-of-state”). Such LSR will help to find other similar relation instances (e.g., 〈Bill Clinton, Madeleine Albright〉). In order to solve the problem, we first define a new meta-path-based relation similarity measure, RelSim, to measure the similarity between relation instances in schema-rich HINs. Then given a query, we propose an optimization model to efficiently learn LSR implied in the query through linear programming, and perform fast relation similarity search using RelSim based on the learned LSR. The experiments on real world datasets derived from Freebase demonstrate the effectiveness and efficiency of our approach.

One of the key obstacles in making learning protocols realistic in applications is the need to supervise them, a costly process that often requires hiring domain experts. We consider the framework to use the world knowledge as indirect supervision. World knowledge is general-purpose knowledge, which is not designed for any specific domain. Then, the key challenges are how to adapt the world knowledge to domains and how to represent it for learning. In this article, we provide an example of using world knowledge for domain-dependent document clustering. We provide three ways to specify the world knowledge to domains by resolving the ambiguity of the entities and their types, and represent the data with world knowledge as a heterogeneous information network. Then, we propose a clustering algorithm that can cluster multiple types and incorporate the sub-type information as constraints. In the experiments, we use two existing knowledge bases as our sources of world knowledge. One is Freebase, which is collaboratively collected knowledge about entities and their organizations. The other is YAGO2, a knowledge base automatically extracted from Wikipedia and maps knowledge to the linguistic knowledge base, WordNet. Experimental results on two text benchmark datasets (20newsgroups and RCV1) show that incorporating world knowledge as indirect supervision can significantly outperform the state-of-the-art clustering algorithms as well as clustering algorithms enhanced with world knowledge features. A preliminary version of this work appeared in the proceedings of KDD 2015 [Wang et al. 2015a]. This journal version has made several major improvements. First, we have proposed a new and general learning framework for machine learning with world knowledge as indirect supervision, where document clustering is a special case in the original paper. Second, in order to make our unsupervised semantic parsing method more understandable, we add several real cases from the original sentences to the resulting logic forms with all the necessary information. Third, we add details of the three semantic filtering methods and conduct deep analysis of the three semantic filters, by using case studies to show why the conceptualization-based semantic filter can produce more accurate indirect supervision. Finally, in addition to the experiment on 20 newsgroup data and Freebase, we have extended the experiments on clustering results by using all the combinations of text (20 newsgroup, MCAT, CCAT, ECAT) and world knowledge sources (Freebase, YAGO2).

With the recent growth of online content on the Web, there have been more user generated data with noisy and missing labels, e.g., social tags and voted labels from Amazon's Mechanical Turks. Most of machine learning methods, which require accurate label sets, could not be trusted when the label sets were yet unreliable. In this paper, we provide a text label refinement algorithm to adjust the labels for such noisy and missing labeled datasets. We assume that the labeled sets can be refined based on the labels with certain confidence, and the similarity between data being consistent with the labels. We propose a label smoothness ratio criterion to measure the smoothness of the labels and the consistency between labels and data. We demonstrate the effectiveness of the label refining algorithm on eight labeled document datasets, and validate that the results are useful for generating better labels.

As a fundamental task, document similarity measure has broad impact to document-based classification, clustering and ranking. Traditional approaches represent documents as bag-of-words and compute document similarities using measures like cosine, Jaccard, and dice. However, entity phrases rather than single words in documents can be critical for evaluating document relatedness. Moreover, types of entities and links between entities/words are also informative. We propose a method to represent a document as a typed heterogeneous information network (HIN), where the entities and relations are annotated with types. Multiple documents can be linked by the words and entities in the HIN. Consequently, we convert the document similarity problem to a graph distance problem. Intuitively, there could be multiple paths between a pair of documents. We propose to use the meta-path defined in HIN to compute distance between documents. Instead of burdening user to define meaningful meta-paths, an automatic method is proposed to rank the meta-paths. Given the meta-paths associated with ranking scores, an HIN-based similarity measure, KnowSim, is proposed to compute document similarities. Using Freebase, a well-known world knowledge base, to conduct semantic parsing and construct HIN for documents, our experiments on 20Newsgroups and RCV1 datasets show that KnowSim generates impressive high-quality document clustering.

One of the key obstacles in making learning protocols realistic in applications is the need to supervise them, a costly process that often requires hiring domain experts. We consider the framework to use the world knowledge as indirect supervision. World knowledge is general-purpose knowledge, which is not designed for any specific domain. Then the key challenges are how to adapt the world knowledge to domains and how to represent it for learning. In this paper, we provide an example of using world knowledge for domain dependent document clustering. We provide three ways to specify the world knowledge to domains by resolving the ambiguity of the entities and their types, and represent the data with world knowledge as a heterogeneous information network. Then we propose a clustering algorithm that can cluster multiple types and incorporate the sub-type information as constraints. In the experiments, we use two existing knowledge bases as our sources of world knowledge. One is Freebase, which is collaboratively collected knowledge about entities and their organizations. The other is YAGO2, a knowledge base automatically extracted from Wikipedia and maps knowledge to the linguistic knowledge base, Word-Net. Experimental results on two text benchmark datasets (20newsgroups and RCV1) show that incorporating world knowledge as indirect supervision can significantly outperform the state-of-the-art clustering algorithms as well as clustering algorithms enhanced with world knowledge features.