Abstract

No Results Found

Abstract

Despite recent advancements showcasing the impressive capabilities of Large Language Models (LLMs) in conversational systems, we show that even state-of-the-art LLMs are morally inconsistent in their generations, questioning their reliability (and trustworthiness in general). Prior works in LLM evaluation focus on developing ground-truth data to measure accuracy on specific tasks. However, for moral scenarios that often lack universally agreed-upon answers, consistency in model responses becomes crucial for their reliability. To address this issue, we propose an information-theoretic measure called Semantic Graph Entropy (SaGE), grounded in the concept of "Rules of Thumb" (RoTs) to measure a model's moral consistency. RoTs are abstract principles learned by a model and can help explain their decision-making strategies effectively. To this extent, we construct the Moral Consistency Corpus (MCC), containing 50K moral questions, responses to them by LLMs, and the RoTs that these models followed. Furthermore, to illustrate the generalizability of SaGE, we use it to investigate LLM consistency on two popular datasets -- TruthfulQA and HellaSwag. Our results reveal that task-accuracy and consistency are independent problems, and there is a dire need to investigate these issues further.

Abstract

A Large Language Model (LLM) is considered consistent if semantically equivalent prompts produce semantically equivalent responses. Despite recent advancements showcasing the impressive capabilities of LLMs in conversational systems, we show that even state-of-the-art LLMs are highly inconsistent in their generations, questioning their reliability. Prior research has tried to measure this with task-specific accuracy. However, this approach is unsuitable for moral scenarios, such as the trolley problem, with no "correct" answer. To address this issue, we propose a novel information-theoretic measure called Semantic Graph Entropy (SGE) to measure the consistency of an LLM in moral scenarios. We leverage "Rules of Thumb" (RoTs) to explain a model's decision-making strategies and further enhance our metric. Compared to existing consistency metrics, SGE correlates better with human judgments across five LLMs. In the future, we aim to investigate the root causes of LLM inconsistencies and propose improvements.

Abstract

Unsupervised Representation Learning on graphs is gaining traction due to the increasing abundance of unlabelled network data and the compactness, richness, and usefulness of the representations generated. In this context, the need to consider fairness and bias constraints while generating the representations has been well-motivated and studied to some extent in prior works. One major limitation of most of the prior works in this setting is that they do not aim to address the bias generated due to connectivity patterns in the graphs, such as varied node centrality, which leads to a disproportionate performance across nodes. In our work, we aim to address this issue of mitigating bias due to inherent graph structure in an unsupervised setting. To this end, we propose CAFIN, a centrality-aware fairness-inducing framework that leverages the structural information of graphs to tune the representations generated by existing frameworks. We deploy it on GraphSAGE (a popular framework in this domain) and showcase its efficacy on two downstream tasks – Node Classification and Link Prediction. Empirically, CAFIN consistently reduces the performance disparity across popular datasets (varying from 18 to 80% reduction in performance disparity) from various domains while incurring only a minimal cost of fairness.

Abstract

No Results Found

Abstract

Machine Learning models face increased concerns regarding the storage of personal user data and adverse impacts of corrupted data like backdoors or systematic bias. Machine Unlearning can address these by allowing post-hoc deletion of affected training data from a learned model. Achieving this task exactly is computationally expensive; consequently, recent works have proposed inexact unlearning algorithms to solve this approximately as well as evaluation methods to test the effectiveness of these algorithms. In this work, we first outline some necessary criteria for evaluation methods and show no existing evaluation satisfies them all. Then, we design a stronger black-box evaluation method called the Interclass Confusion (IC) test which adversarially manipulates data during training to detect the insufficiency of unlearning procedures. We also propose two analytically motivated baseline methods~(EU-k and CF-k) which outperform several popular inexact unlearning methods. Overall, we demonstrate how adversarial evaluation strategies can help in analyzing various unlearning phenomena which can guide the development of stronger unlearning algorithms.

Abstract

Prior work has shown that pretrained language models often make incorrect predictions for negated inputs. The reason for this behaviour has remained unclear. It has been argued that since language models (LMs) don’t change their predictions about factual propositions under negation, they might not detect negation. We show encoder LMs do detect negation as their representations across layers reliably distinguish negated inputs from non-negated inputs, and when negation leads to contradictions. However, probing experiments show that these LMs indeed don’t use negation when evaluating whether a factual statement is true, even when fine-tuned with the objective of changing outputs on negated sentences (Hosseini et al., 2021). We hypothesize about why pretrained LMs are inconsistent under negation: when the statement could refer to multiple ground entities with conflicting properties, negation may not entail a change in output. This means negation minimal pairs in different training samples can have the same completion in pretraining corpora. We argue pretraining may not provide enough signal to learn the distribution of ground referents a token could have, confusing the LM on how to handle negation.

Abstract

No Results Found

Abstract

Social media provides a supportive and anonymous environment for discussing mental health issues, including depression. Existing research on identifying the cause of depression focuses primarily on improving classifier models, while neglecting the importance of learning better data representations. To address this gap, we introduce an architecture that enhances the identification of the cause of depression by learning improved data representations. Our work enables a deeper interpretation of the cause of depression in social media contexts, emphasizing the significance of effective rep- resentation learning for this task. Our work can act as a foundation for self-help applications in the field of mental health.

Abstract

The default approach to deal with the enormous size and lim- ited accessibility of many Web and social media networks is to sample one or more subnetworks from a conceptually unbounded unknown network. Clearly, the extracted subnet- works will crucially depend on the sampling scheme. Mo- tivated by studies of homophily and opinion formation, we propose a variant of snowball sampling designed to priori- tize inclusion of entire cohesive communities rather than any kind of representativeness, breadth, or depth of coverage. The method is illustrated on a concrete example, and experiments on synthetic networks suggest that it behaves as desired.