ML models often exhibit unexpectedly poor behavior when they are deployed in real-world domains. We identify underspecification as a key reason for these failures. An ML pipeline is underspecified when it can return many predictors with equivalently strong held-out performance in the training domain. Underspecification is common in modern ML pipelines, such as those based on deep learning. Predictors returned by underspecified pipelines are often treated as equivalent based on their training domain performance, but we show here that such predictors can behave very differently in deployment domains. This ambiguity can lead to instability and poor model behavior in practice, and is a distinct failure mode from previously identified issues arising from structural mismatch between training and deployment domains. We show that this problem appears in a wide variety of practical ML pipelines, using examples from computer vision, medical imaging, natural language processing, clinical risk prediction based on electronic health records, and medical genomics. Our results show the need to explicitly account for underspecification in modeling pipelines that are intended for real-world deployment in any domain.

Underspecification Presents Challenges for Credibility in Modern Machine Learning

Data-driven decision-making consequential to individuals raises important questions of accountability and justice. Indeed, European law provides individuals limited rights to 'meaningful information about the logic' behind significant, autonomous decisions such as loan approvals, insurance quotes, and CV filtering. We undertake three experimental studies examining people's perceptions of justice in algorithmic decision-making under different scenarios and explanation styles. Dimensions of justice previously observed in response to human decision-making appear similarly engaged in response to algorithmic decisions. Qualitative analysis identified several concerns and heuristics involved in justice perceptions including arbitrariness, generalisation, and (in)dignity. Quantitative analysis indicates that explanation styles primarily matter to justice perceptions only when subjects are exposed to multiple different styles --- under repeated exposure of one style, scenario effects obscure any explanation effects. Our results suggests there may be no 'best' approach to explaining algorithmic decisions, and that reflection on their automated nature both implicates and mitigates justice dimensions.

'It's Reducing a Human Being to a Percentage': Perceptions of Justice in Algorithmic Decisions

Although recent research has shown that marine litter has made it even to the remotest parts of our planet, little information is available about temporal trends on the deep ocean floor. To quantify litter on the deep seafloor over time, we analysed images from the HAUSGARTEN observatory (79°N) taken in 2002, 2004, 2007, 2008 and 2011 (2500 m depth). Our results indicate that litter increased from 3635 to 7710 items km⁻² between 2002 and 2011 and reached densities similar to those reported from a canyon near the Portuguese capital Lisboa. Plastic constituted the majority of litter (59%) followed by a black fabric (11%) and cardboard/paper (7%). Sixty-seven percent of the litter was entangled or colonised by invertebrates such as sponges (41%) or sea anemones (15%). The changes in litter could be an indirect consequence of the receding sea ice, which opens the Arctic Ocean to the impacts of man's activities.

Increase of litter at the Arctic deep-sea observatory HAUSGARTEN – indirect effect of global warming?

Interpretability in machine learning (ML) is crucial for high stakes decisions and troubleshooting. In this work, we provide fundamental principles for interpretable ML, and dispel common misunderstandings that dilute the importance of this crucial topic. We also identify 10 technical challenge areas in interpretable machine learning and provide history and background on each problem. Some of these problems are classically important, and some are recent problems that have arisen in the last few years. These problems are: (1) Optimizing sparse logical models such as decision trees; (2) Optimization of scoring systems; (3) Placing constraints into generalized additive models to encourage sparsity and better interpretability; (4) Modern case-based reasoning, including neural networks and matching for causal inference; (5) Complete supervised disentanglement of neural networks; (6) Complete or even partial unsupervised disentanglement of neural networks; (7) Dimensionality reduction for data visualization; (8) Machine learning models that can incorporate physics and other generative or causal constraints; (9) Characterization of the "Rashomon set" of good models; and (10) Interpretable reinforcement learning. This survey is suitable as a starting point for statisticians and computer scientists interested in working in interpretable machine learning.

Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges

Biomass waste-derived porous carbons (BWDPCs) are a class of complex materials that are widely used in sustainable waste management and carbon capture. However, their diverse textural properties, the presence of various functional groups, and the varied temperatures and pressures to which they are subjected during CO2 adsorption make it challenging to understand the underlying mechanism of CO2 adsorption. Here, we compiled a data set including 527 data points collected from peer-reviewed publications and applied machine learning to systematically map CO2 adsorption as a function of the textural and compositional properties of BWDPCs and adsorption parameters. Various tree-based models were devised, where the gradient boosting decision trees (GBDTs) had the best predictive performance with R2 of 0.98 and 0.84 on the training and test data, respectively. Further, the BWDPCs in the compiled data set were classified into regular porous carbons (RPCs) and heteroatom-doped porous carbons (HDPCs), where again the GBDT model had R2 of 0.99 and 0.98 on the training and 0.86 and 0.79 on the test data for the RPCs and HDPCs, respectively. Feature importance revealed the significance of adsorption parameters, textural properties, and compositional properties in the order of precedence for BWDPC-based CO2 adsorption, effectively guiding the synthesis of porous carbons for CO2 adsorption applications.

Applied Machine Learning for Prediction of CO2 Adsorption on Biomass Waste-Derived Porous Carbons.

To advance our understanding of the environmental fate and transport of macro- and micro-plastic debris, robust and reproducible methods, technologies, and analytical approaches are necessary for in situ plastic-type identification and characterization. This investigation compares four spectroscopic techniques: attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR), near-infrared (NIR) reflectance spectroscopy, laser-induced breakdown spectroscopy (LIBS), and X-ray fluorescence (XRF) spectroscopy, coupled to seven classification methods, including machine learning classifiers, to determine accuracy for identifying type of both consumer plastics and marine plastic debris (MPD). With machine learning classifiers, consumer plastic types were identified with 99, 91, 97, and 70% success rates for ATR-FTIR, NIR reflectance spectroscopy, LIBS, and XRF, respectively. The classification of MPD had similar or lower success rates, likely arising from alterations to the plastic from environmental weathering processes with success rates of 99, 81, 76, and 66% for ATR-FTIR, NIR reflectance spectroscopy, LIBS, and XRF, respectively. Success rates indicate that ATR-FTIR, NIR reflectance spectroscopy, and LIBS coupled with machine learning classifiers can be used to identify both consumer and environmental plastic samples.

Rapid Identification of Marine Plastic Debris via Spectroscopic Techniques and Machine Learning Classifiers.

Prediction problems often admit competing models that perform almost equally well. This effect challenges key assumptions in machine learning when competing models assign conflicting predictions. In this paper, we define predictive multiplicity as the ability of a prediction problem to admit competing models with conflicting predictions. We introduce formal measures to evaluate the severity of predictive multiplicity and develop integer programming tools to compute them exactly for linear classification problems. We apply our tools to measure predictive multiplicity in recidivism prediction problems. Our results show that real-world datasets may admit competing models that assign wildly conflicting predictions, and motivate the need to measure and report predictive multiplicity in model development.

Predictive Multiplicity in Classification

Motivated by the need to audit complex and black box models, there has been extensive research on quantifying how data features influence model predictions. Feature influence can be direct (a direct influence on model outcomes) and indirect (model outcomes are influenced via proxy features). Feature influence can also be expressed in aggregate over the training or test data or locally with respect to a single point. Current research has typically focused on one of each of these dimensions. In this paper, we develop disentangled influence audits, a procedure to audit the indirect influence of features. Specifically, we show that disentangled representations provide a mechanism to identify proxy features in the dataset, while allowing an explicit computation of feature influence on either individual outcomes or aggregate-level outcomes. We show through both theory and experiments that disentangled influence audits can both detect proxy features and show, for each individual or in aggregate, which of these proxy features affects the classifier being audited the most. In this respect, our method is more powerful than existing methods for ascertaining feature influence.

/pdf/disentangling-influence-using-disentangled-representations-42jghjikxn.pdf

Disentangling Influence: Using disentangled representations to audit model predictions

/pdf/predictive-multiplicity-in-classification-1401mshw4u.pdf

Charles T. Marx

Papers

Rapid Identification of Marine Plastic Debris via Spectroscopic Techniques and Machine Learning Classifiers.

Predictive Multiplicity in Classification

Disentangling Influence: Using disentangled representations to audit model predictions

Predictive Multiplicity in Classification

Disentangling Influence: Using Disentangled Representations to Audit Model Predictions