Multi-modal Large Language Models (MLLMs) have shown impressive abilities in generating reasonable responses with respect to multi-modal contents. However, there is still a wide gap between the performance of recent MLLM-based applications and the expectation of the broad public, even though the most powerful OpenAI's GPT-4 and Google's Gemini have been deployed. This paper strives to enhance understanding of the gap through the lens of a qualitative study on the generalizability, trustworthiness, and causal reasoning capabilities of recent proprietary and open-source MLLMs across four modalities: ie, text, code, image, and video, ultimately aiming to improve the transparency of MLLMs. We believe these properties are several representative factors that define the reliability of MLLMs, in supporting various downstream applications. To be specific, we evaluate the closed-source GPT-4 and Gemini and 6 open-source LLMs and MLLMs. Overall we evaluate 230 manually designed cases, where the qualitative results are then summarized into 12 scores (ie, 4 modalities times 3 properties). In total, we uncover 14 empirical findings that are useful to understand the capabilities and limitations of both proprietary and open-source MLLMs, towards more reliable downstream multi-modal applications.
Much of the world's most valued data is stored in relational databases and data warehouses, where the data is organized into many tables connected by primary-foreign key relations. However, building machine learning models using this data is both challenging and time consuming. The core problem is that no machine learning method is capable of learning on multiple tables interconnected by primary-foreign key relations. Current methods can only learn from a single table, so the data must first be manually joined and aggregated into a single training table, the process known as feature engineering. Feature engineering is slow, error prone and leads to suboptimal models. Here we introduce an end-to-end deep representation learning approach to directly learn on data laid out across multiple tables. We name our approach Relational Deep Learning (RDL). The core idea is to view relational databases as a temporal, heterogeneous graph, with a node for each row in each table, and edges specified by primary-foreign key links. Message Passing Graph Neural Networks can then automatically learn across the graph to extract representations that leverage all input data, without any manual feature engineering. Relational Deep Learning leads to more accurate models that can be built much faster. To facilitate research in this area, we develop RelBench, a set of benchmark datasets and an implementation of Relational Deep Learning. The data covers a wide spectrum, from discussions on Stack Exchange to book reviews on the Amazon Product Catalog. Overall, we define a new research area that generalizes graph machine learning and broadens its applicability to a wide set of AI use cases.
The growing awareness of safety concerns in large language models (LLMs) has sparked considerable interest in the evaluation of safety within current research endeavors. This study investigates an interesting issue pertaining to the evaluation of LLMs, namely the substantial discrepancy in performance between multiple-choice questions and open-ended questions. Inspired by research on jailbreak attack patterns, we argue this is caused by mismatched generalization. That is, the LLM does not have a comprehensive understanding of the complex concept of safety. Instead, it only remembers what to answer for open-ended safety questions, which makes it unable to solve other forms of safety tests. We refer to this phenomenon as fake alignment and construct a comparative benchmark to empirically verify its existence in LLMs. Such fake alignment renders previous evaluation protocols unreliable. To address this, we introduce the Fake alIgNment Evaluation (FINE) framework and two novel metrics--Consistency Score (CS) and Consistent Safety Score (CSS), which jointly assess two complementary forms of evaluation to quantify fake alignment and obtain corrected performance estimates. Applying FINE to 14 widely-used LLMs reveals several models with purported safety are poorly aligned in practice. Our work highlights potential limitations in prevailing alignment methodologies.
The widespread adoption of large language models (LLMs) across various regions underscores the urgent need to evaluate their alignment with human values. Current benchmarks, however, fall short of effectively uncovering safety vulnerabilities in LLMs. Despite numerous models achieving high scores and 'topping the chart' in these evaluations, there is still a significant gap in LLMs' deeper alignment with human values and achieving genuine harmlessness. To this end, this paper proposes the first highly adversarial benchmark named Flames, consisting of 2,251 manually crafted prompts, ~18.7K model responses with fine-grained annotations, and a specified scorer. Our framework encompasses both common harmlessness principles, such as fairness, safety, legality, and data protection, and a unique morality dimension that integrates specific Chinese values such as harmony. Based on the framework, we carefully design adversarial prompts that incorporate complex scenarios and jailbreaking methods, mostly with implicit malice. By prompting mainstream LLMs with such adversarially constructed prompts, we obtain model responses, which are then rigorously annotated for evaluation. Our findings indicate that all the evaluated LLMs demonstrate relatively poor performance on Flames, particularly in the safety and fairness dimensions. Claude emerges as the best-performing model overall, but with its harmless rate being only 63.08% while GPT-4 only scores 39.04%. The complexity of Flames has far exceeded existing benchmarks, setting a new challenge for contemporary LLMs and highlighting the need for further alignment of LLMs. To efficiently evaluate new models on the benchmark, we develop a specified scorer capable of scoring LLMs across multiple dimensions, achieving an accuracy of 77.4%. The Flames Benchmark is publicly available on https://github.com/AIFlames/Flames.
Offline reinforcement learning (RL), a technology that offline learns a policy from logged data without the need to interact with online environments, has become a favorable choice in decision-making processes like interactive recommendation. Offline RL faces the value overestimation problem. To address it, existing methods employ conservatism, e.g., by constraining the learned policy to be close to behavior policies or punishing the rarely visited state-action pairs. However, when applying such offline RL to recommendation, it will cause a severe Matthew effect, i.e., the rich get richer and the poor get poorer, by promoting popular items or categories while suppressing the less popular ones. It is a notorious issue that needs to be addressed in practical recommender systems. In this paper, we aim to alleviate the Matthew effect in offline RL-based recommendation. Through theoretical analyses, we find that the conservatism of existing methods fails in pursuing users' long-term satisfaction. It inspires us to add a penalty term to relax the pessimism on states with high entropy of the logging policy and indirectly penalizes actions leading to less diverse states. This leads to the main technical contribution of the work: Debiased model-based Offline RL (DORL) method. Experiments show that DORL not only captures user interests well but also alleviates the Matthew effect. The implementation is available via https://github.com/chongminggao/DORL-codes.
Machine learning models exhibit strong performance on datasets with abundant labeled samples. However, for tabular datasets with extremely high $d$-dimensional features but limited $n$ samples (i.e. $d \gg n$), machine learning models struggle to achieve strong performance due to the risk of overfitting. Here, our key insight is that there is often abundant, auxiliary domain information describing input features which can be structured as a heterogeneous knowledge graph (KG). We propose PLATO, a method that achieves strong performance on tabular data with $d \gg n$ by using an auxiliary KG describing input features to regularize a multilayer perceptron (MLP). In PLATO, each input feature corresponds to a node in the auxiliary KG. In the MLP's first layer, each input feature also corresponds to a weight vector. PLATO is based on the inductive bias that two input features corresponding to similar nodes in the auxiliary KG should have similar weight vectors in the MLP's first layer. PLATO captures this inductive bias by inferring the weight vector for each input feature from its corresponding node in the KG via a trainable message-passing function. Across 6 $d \gg n$ datasets, PLATO outperforms 13 state-of-the-art baselines by up to 10.19%.
Graph Neural Networks (GNNs) are powerful machine learning prediction models on graph-structured data. However, GNNs lack rigorous uncertainty estimates, limiting their reliable deployment in settings where the cost of errors is significant. We propose conformalized GNN (CF-GNN), extending conformal prediction (CP) to graph-based models for guaranteed uncertainty estimates. Given an entity in the graph, CF-GNN produces a prediction set/interval that provably contains the true label with pre-defined coverage probability (e.g. 90%). We establish a permutation invariance condition that enables the validity of CP on graph data and provide an exact characterization of the test-time coverage. Moreover, besides valid coverage, it is crucial to reduce the prediction set size/interval length for practical use. We observe a key connection between non-conformity scores and network structures, which motivates us to develop a topology-aware output correction model that learns to update the prediction and produces more efficient prediction sets/intervals. Extensive experiments show that CF-GNN achieves any pre-defined target marginal coverage while significantly reducing the prediction set/interval size by up to 74% over the baselines. It also empirically achieves satisfactory conditional coverage over various raw and network features.
Interactive data exploration (IDE) is an effective way of comprehending big data, whose volume and complexity are beyond human abilities. The main goal of IDE is to discover user interest regions from a database through multi-rounds of user labelling. Existing IDEs adopt active-learning framework, where users iteratively discriminate or label the interestingness of selected tuples. The process of data exploration can be viewed as the process of training a classifier, which determines whether a database tuple is interesting to a user. An efficient exploration thus takes very few iterations of user labelling to reach the data region of interest. In this work, we consider the data exploration as the process of few-shot learning, where the classifier is learned with only a few training examples, or exploration iterations. To this end, we propose a learning-to-explore framework, based on meta-learning, which learns how to learn a classifier with automatically generated meta-tasks, so that the exploration process can be much shortened. Extensive experiments on real datasets show that our proposal outperforms existing explore-by-example solutions in terms of accuracy and efficiency.
Despite many advances in Graph Neural Networks (GNNs), their training strategies simply focus on minimizing a loss over nodes in a graph. However, such simplistic training strategies may be sub-optimal as they neglect that certain nodes are much harder to make accurate predictions on than others. Here we present TuneUp, a curriculum learning strategy for better training GNNs. Crucially, TuneUp trains a GNN in two stages. The first stage aims to produce a strong base GNN. Such base GNNs tend to perform well on head nodes (nodes with large degrees) but less so on tail nodes (nodes with small degrees). So, the second stage of TuneUp specifically focuses on improving prediction on tail nodes. Concretely, TuneUp synthesizes many additional supervised tail node data by dropping edges from head nodes and reusing the supervision on the original head nodes. TuneUp then minimizes the loss over the synthetic tail nodes to finetune the base GNN. TuneUp is a general training strategy that can be used with any GNN architecture and any loss, making TuneUp applicable to a wide range of prediction tasks. Extensive evaluation of TuneUp on two GNN architectures, three types of prediction tasks, and both inductive and transductive settings shows that TuneUp significantly improves the performance of the base GNN on tail nodes, while often even improving the performance on head nodes, which together leads up to 58.5% relative improvement in GNN predictive performance. Moreover, TuneUp significantly outperforms its variants without the two-stage curriculum learning, existing graph data augmentation techniques, as well as other specialized methods for tail nodes.