To develop rigorous knowledge about ML models -- and the systems in which they are embedded -- we need reliable measurements. But reliable measurement is fundamentally challenging, and touches on issues of reproducibility, scalability, uncertainty quantification, epistemology, and more. This dissertation addresses criteria needed to take reliability seriously: both criteria for designing meaningful metrics, and for methodologies that ensure that we can dependably and efficiently measure these metrics at scale and in practice. In doing so, this dissertation articulates a research vision for a new field of scholarship at the intersection of machine learning, law, and policy. Within this frame, we cover topics that fit under three different themes: (1) quantifying and mitigating sources of arbitrariness in ML, (2) taming randomness in uncertainty estimation and optimization algorithms, in order to achieve scalability without sacrificing reliability, and (3) providing methods for evaluating generative-AI systems, with specific focuses on quantifying memorization in language models and training latent diffusion models on open-licensed data. By making contributions in these three themes, this dissertation serves as an empirical proof by example that research on reliable measurement for machine learning is intimately and inescapably bound up with research in law and policy. These different disciplines pose similar research questions about reliable measurement in machine learning. They are, in fact, two complementary sides of the same research vision, which, broadly construed, aims to construct machine-learning systems that cohere with broader societal values.

Vision-Language Models (VLMs) are becoming increasingly vulnerable to adversarial attacks as various novel attack strategies are being proposed against these models. While existing defenses excel in unimodal contexts, they currently fall short in safeguarding VLMs against adversarial threats. To mitigate this vulnerability, we propose a novel, yet elegantly simple approach for detecting adversarial samples in VLMs. Our method leverages Text-to-Image (T2I) models to generate images based on captions produced by target VLMs. Subsequently, we calculate the similarities of the embeddings of both input and generated images in the feature space to identify adversarial samples. Empirical evaluations conducted on different datasets validate the efficacy of our approach, outperforming baseline methods adapted from image classification domains. Furthermore, we extend our methodology to classification tasks, showcasing its adaptability and model-agnostic nature. Theoretical analyses and empirical findings also show the resilience of our approach against adaptive attacks, positioning it as an excellent defense mechanism for real-world deployment against adversarial threats.

We present the findings of the first NeurIPS competition on unlearning, which sought to stimulate the development of novel algorithms and initiate discussions on formal and robust evaluation methodologies. The competition was highly successful: nearly 1,200 teams from across the world participated, and a wealth of novel, imaginative solutions with different characteristics were contributed. In this paper, we analyze top solutions and delve into discussions on benchmarking unlearning, which itself is a research problem. The evaluation methodology we developed for the competition measures forgetting quality according to a formal notion of unlearning, while incorporating model utility for a holistic evaluation. We analyze the effectiveness of different instantiations of this evaluation framework vis-a-vis the associated compute cost, and discuss implications for standardizing evaluation. We find that the ranking of leading methods remains stable under several variations of this framework, pointing to avenues for reducing the cost of evaluation. Overall, our findings indicate progress in unlearning, with top-performing competition entries surpassing existing algorithms under our evaluation framework. We analyze trade-offs made by different algorithms and strengths or weaknesses in terms of generalizability to new datasets, paving the way for advancing both benchmarking and algorithm development in this important area.

Network intrusion detection is one of the most important issues in the field of cyber security, and various machine learning techniques have been applied to build intrusion detection systems. However, since the number of features to describe the network connections is often large, where some features are redundant or noisy, feature selection is necessary in such scenarios, which can both improve the efficiency and accuracy. Recently, some researchers focus on using multi-objective evolutionary algorithms (MOEAs) to select features. But usually, they only consider the number of features and classification accuracy as the objectives, resulting in unsatisfactory performance on a critical metric, detection rate. This will lead to the missing of many real attacks and bring huge losses to the network system. In this paper, we propose DR-MOFS to model the feature selection problem in network intrusion detection as a three-objective optimization problem, where the number of features, accuracy and detection rate are optimized simultaneously, and use MOEAs to solve it. Experiments on two popular network intrusion detection datasets NSL-KDD and UNSW-NB15 show that in most cases the proposed method can outperform previous methods, i.e., lead to fewer features, higher accuracy and detection rate.

Large language models (LLMs) have demonstrated potential in the innovation of many disciplines. However, how they can best be developed for oncology remains underdeveloped. State-of-the-art OpenAI models were fine-tuned on a clinical dataset and clinical guidelines text corpus for two important cancer treatment factors, adjuvant radiation therapy and chemotherapy, using a novel Langchain prompt engineering pipeline. A high accuracy (0.85+) was achieved in the classification of adjuvant radiation therapy and chemotherapy for breast cancer patients. Furthermore, a confidence interval was formed from observational data on the quality of treatment from human oncologists to estimate the proportion of scenarios in which the model must outperform the original oncologist in its treatment prediction to be a better solution overall as 8.2% to 13.3%. Due to indeterminacy in the outcomes of cancer treatment decisions, future investigation, potentially a clinical trial, would be required to determine if this threshold was met by the models. Nevertheless, with 85% of U.S. cancer patients receiving treatment at local community facilities, these kinds of models could play an important part in expanding access to quality care with outcomes that lie, at minimum, close to a human oncologist.

In silico methods evaluating the mutation effects of missense mutations are providing an important approach for understanding mutations in personal genomes and identifying disease-relevant biomarkers. However, existing methods, including deep learning methods, heavily rely on sequence-aware information, and do not fully leverage the potential of available 3D structural information. In addition, these methods may exhibit an inability to predict mutations in domains difficult to formulate sequence-based embeddings. In this study, we introduce a novel rationale-guided graph neural network AlphaGMut to evaluate mutation effects and to distinguish pathogenic mutations from neutral mutations. We compute the alpha shapes of protein structures to obtain atomic-resolution edge connectivities and map them to an accurate residue-level graph representation. We then compute structural-, topological-, biophysical-, and sequence properties of the mutation sites, which are assigned as node attributes in the graph. These node attributes could effectively guide the graph neural network to learn the difference between pathogenic and neutral mutations using k-hop message passing with a short training period. We demonstrate that AlphaGMut outperforms state-of-the-art methods, including DeepMind's AlphaMissense, in many performance metrics. In addition, AlphaGMut has the advantage of performing well in alignment-free settings, which provides broader prediction coverage and better generalization compared to current methods requiring deep sequence-aware information.

This study assesses the ability of Large Vision-Language Models (LVLMs) to differentiate between AI-generated and human-generated images. It introduces a new automated benchmark construction method for this evaluation. The experiment compared common LVLMs with human participants using a mixed dataset of AI and human-created images. Results showed that LVLMs could distinguish between the image types to some extent but exhibited a rightward bias, and perform significantly worse compared to humans. To build on these findings, we developed an automated benchmark construction process using AI. This process involved topic retrieval, narrative script generation, error embedding, and image generation, creating a diverse set of text-image pairs with intentional errors. We validated our method through constructing two caparable benchmarks. This study highlights the strengths and weaknesses of LVLMs in real-world understanding and advances benchmark construction techniques, providing a scalable and automatic approach for AI model evaluation.

Electronic healthcare records are vital for patient safety as they document conditions, plans, and procedures in both free text and medical codes. Language models have significantly enhanced the processing of such records, streamlining workflows and reducing manual data entry, thereby saving healthcare providers significant resources. However, the black-box nature of these models often leaves healthcare professionals hesitant to trust them. State-of-the-art explainability methods increase model transparency but rely on human-annotated evidence spans, which are costly. In this study, we propose an approach to produce plausible and faithful explanations without needing such annotations. We demonstrate on the automated medical coding task that adversarial robustness training improves explanation plausibility and introduce AttInGrad, a new explanation method superior to previous ones. By combining both contributions in a fully unsupervised setup, we produce explanations of comparable quality, or better, to that of a supervised approach. We release our code and model weights.

Efficient and current roadway geometry data collection is a critical task for transportation agencies to undertake effective road planning, maintenance, design, and rehabilitation efforts. The methods for gathering such data can be broadly classified into two categories: a) land-based methods, which encompass field inventory, mobile mapping, and image logging, and b) aerial-based methods, which involve satellite imagery, drones, and laser scanning. However, employing land-based techniques for extensive highway networks covering thousands of miles proves arduous and costly, and poses safety risks for crew members. Consequently, there exists a pressing need to develop more efficient methodologies for acquiring this data promptly, safely, and economically. Fortunately, with the increasing availability of high-resolution images and recent strides in computer vision and object detection technologies, automated extraction of roadway geometry features has become feasible.

The fairness of AI decision-making has garnered increasing attention, leading to the proposal of numerous fairness algorithms. In this paper, we aim not to address this issue by directly introducing fair learning algorithms, but rather by generating entirely new, fair synthetic data from biased datasets for use in any downstream tasks. Additionally, the distribution of test data may differ from that of the training set, potentially impacting the performance of the generated synthetic data in downstream tasks. To address these two challenges, we propose a diffusion model-based framework, FADM: Fairness-Aware Diffusion with Meta-training. FADM introduces two types of gradient induction during the sampling phase of the diffusion model: one to ensure that the generated samples belong to the desired target categories, and another to make the sensitive attributes of the generated samples difficult to classify into any specific sensitive attribute category. To overcome data distribution shifts in the test environment, we train the diffusion model and the two classifiers used for induction within a meta-learning framework. Compared to other baselines, FADM allows for flexible control over the categories of the generated samples and exhibits superior generalization capability. Experiments on real datasets demonstrate that FADM achieves better accuracy and optimal fairness in downstream tasks.