Meaningful insights: explainability techniques for black-box models on tabular data

Artificial Intelligence (AI) and Machine Learning (ML) are novel data analysis techniques providing very accurate prediction results. They are widely adopted in a variety of industries to improve efficiency and decision-making, but they are also being used to develop intelligent systems. Their success grounds upon complex mathematical models, whose decisions and rationale are usually difficult to comprehend for human users to the point of being dubbed as black-boxes. This is particularly relevant in sensitive and highly regulated domains. To mitigate and possibly solve this issue, the Explainable AI (XAI) field became prominent in recent years. XAI consists of models and techniques to enable understanding of the intricated patterns discovered by black-box models. In this thesis, we consider model-agnostic XAI techniques, which can be applied to Tabular data, with a particular focus on the Credit Scoring domain. Special attention is dedicated to the LIME framework, for which we propose several modifications to the vanilla algorithm, in particular: a pair of complementary Stability Indices that accurately measure LIME stability, and the OptiLIME policy which helps the practitioner finding the proper balance among explanations' stability and reliability. We subsequently put forward GLEAMS a model-agnostic surrogate interpretable model which requires to be trained only once, while providing both Local and Global explanations of the black-box model. GLEAMS produces feature attributions and what-if scenarios, from both dataset and model perspective. Eventually, we argue that synthetic data are an emerging trend in AI, being more and more used to train complex models instead of original data. To be able to explain the outcomes of such models, we must guarantee that synthetic data are reliable enough to be able to translate their explanations to real-world individuals. To this end we propose DAISYnt, a suite of tests to measure synthetic tabular data quality and privacy.

Regulation of Port Charges in Spain: Global versus Local Competition

This article examines the determinants of traffic volumes and the revenues per tonne generated by Spain’s port authorities. The interest of the study lies on the strong differences between port authorities in a context of strict regulation but that provides some scope for price competition. We find that port charges influence the amount of traffic that a port is able to generate. Furthermore, we find clear evidence of local price competition and report mixed results for global competition. Revenues per tonne are higher in ports operating more international regular lines and with multinational terminal operators, while they are lower in ports with nearby competing facilities and where the market share of the dominant shipping firm is high.

Governing the clean development mechanism: global rhetoric versus local realities in carbon sequestration projects

Global agreements have proliferated in the past ten years. One of these is the Kyoto Protocol, which contains provisions for emissions reductions by trading carbon through the Clean Development Mechanism (CDM). The CDM is a market-based instrument that allows companies in Annex I countries to offset their greenhouse gas emissions through energy and tree offset projects in the global South. I set out to examine the governance challenges posed by the institutional design of carbon sequestration projects under the CDM. I examine three global narratives associated with the design of CDM forest projects, specifically North – South knowledge politics, green developmentalism, and community participation, and subsequently assess how these narratives match with local practices in two projects in Latin America. Findings suggest that governance problems are operating at multiple levels and that the rhetoric of global carbon actors often asserts these schemes in one light, while the rhetoric of those who are immediately involved locally may be different. I also stress the alarmist’s discourse that blames local people for the problems of environmental change. The case studies illustrate the need for vertical communication and interaction and nested governance arrangements as well as horizontal arrangements. I conclude that the global framing of forests as offsets requires better integration of local relationships to forests and their management and more effective institutions at multiple levels to link the very local to the very large scale when dealing with carbon sequestration in the CDM.

Local ownership and UN peacebuilding: discourse versus operationalization

The UN asserts that local ownership boosts the legitimacy and sustainabil- ity of peacebuilding by preserving the principles of self-determination and nonimposition in an activity that can contravene them. At the same time, it also perceives local ownership to imperil the achievement of its operational goals, thus bringing its normative and operational obligations into conflict. This article evaluates the UN’s discourse and operationalization of local ownership, showing that despite the UN’s invocation of ownership dis- course, it operationalizes ownership in restrictive ways that are intended to protect the achievement of operational goals but that consequently limit self-determination and increase imposition. Moreover, because of contra- dictions in the UN’s practices of ownership, it also undercuts its ability to re- alize the very operational goals that it is trying to protect. KEYWORDS: UN peacebuilding, local ownership, discourse vs. practice.

Part of the global DOC versus AOU (dissolved organic carbon/apparent oxygen utilization) data compilation, Arabian Sea

Concentrations of total organic carbon (TOC) were determined on samples collected during six cruises in the northern Arabian Sea during the 1995 US JGOFS Arabian Sea Process Study. Total organic carbon concentrations and integrated stocks in the upper ocean varied both spatially and seasonally. Highest mixed-layer TOC concentrations (80-100 µM C) were observed near the coast when upwelling was not active, while upwelling tended to reduce local concentrations. In the open ocean, highest mixed-layer TOC concentrations (80-95 µM C) developed in winter (period of the NE Monsoon) and remained through mid summer (early to mid-SW Monsoon). Lowest open ocean mixed-layer concentrations (65-75 µM C) occurred late in the summer (late SW Monsoon) and during the Fall Intermonsoon period. The changes in TOC concentrations resulted in seasonal variations in mean TOC stocks (upper 150 m) of 1.5-2 mole C/m**2, with the lowest stocks found late in the summer during the SW Monsoon-Fall Intermonsoon transition. The seasonal accumulation of TOC north of 15°N was 31-41 x 10**12 g C, mostly taking place over the period of the NE Monsoon, and equivalent to 6-8% of annual primary production estimated for that region in the mid-1970s. A net TOC production rate of 12 mmole C/m**2/d over the period of the NE Monsoon represented ~80% of net community production. Net TOC production was nil during the SW Monsoon, so vertical export would have dominated the export terms over that period. Total organic carbon concentrations varied in vertical profiles with the vertical layering of the water masses, with the Persian Gulf Water TOC concentrations showing a clear signal. Deep water (>2000 m) TOC concentrations were uniform across the basin and over the period of the cruises, averaging 42.3±1.4 µM C.

Part of the global DOC versus AOU (dissolved organic carbon/apparent oxygen utilization) data compilation, Azores-I

Temperate, transitional and subtropical waters of the remote Azores Front region east of Azores (24-40°N, 22-32°W) were sampled during three cruises conducted under increasing stratification conditions (April 1999, May 1997 and August 1998). Despite the temporal increase of surface temperature (by 5 °C) and stratification (by 2.1 1/min**2), as well as the thermocline shoaling (by ~15 m), dissolved organic carbon (DOC) and nitrogen (DON) in the surface layer were not significantly different for the early spring, late spring and summer periods, with average concentrations of 69±2 µM-C and 5.2±0.4 µM-N, respectively. The surface excess of semi-labile DOC, compared with the baseline DOC concentration in the deep ocean (47±2 µM-C), represents 33% of the bulk DOC concentration and as much as 85% of the TOC (=POC+DOC) excess. When compared with the winter baseline (56±2 µM-C), the seasonal surface DOC excess is 20% of the bulk DOC concentration and 87% of the seasonal TOC excess. These results confirm the major role played by DOC in the carbon cycle of surface waters of the Azores Front region. The total amount of bioreactive DOC transported from the temperate to the subtropical North Atlantic by the Ekman flux between March and December represents only ~15% of the average annual primary production, and ~15% and ~30% of the measured sinking POC flux+vertical DOC eddy diffusion during early spring and summer, respectively. Vertical eddy diffusion is 35% and 2% of the spring and summer sinking POC flux, respectively. On the other hand, DOC only contributes 13% to the local oxidation of organic matter in subsurface waters (between the pycnocline and 500 m) of the study region.

Residual delay maps unveil global patterns of atmospheric nonlinearity and produce improved local forecasts

We use residual-delay maps of observational field data for barometric pressure to demonstrate the structure of latitudinal gradients in nonlinearity in the atmosphere. Nonlinearity is weak and largely lacking in tropical and subtropical sites and increases rapidly into the temperate regions where the time series also appear to be much noisier. The degree of nonlinearity closely follows the meridional variation of midlatitude storm track frequency. We extract the specific functional form of this nonlinearity, a V shape in the lagged residuals that appears to be a basic feature of midlatitude synoptic weather systems associated with frontal passages. We present evidence that this form arises from the relative time scales of high-pressure versus low-pressure events. Finally, we show that this nonlinear feature is weaker in a well regarded numerical forecast model (European Centre for Medium-Range Forecasts) because small-scale temporal and spatial variation is smoothed out in the grided inputs. This is significant, in that it allows us to demonstrate how application of statistical corrections based on the residual-delay map may provide marked increases in local forecast accuracy, especially for severe weather systems.

Layout and material gradation in topology optimization of functionally graded structures: a global-local approach

By means of continuous topology optimization, this paper discusses the influence of material gradation and layout in the overall stiffness behavior of functionally graded structures. The formulation is associated to symmetry and pattern repetition constraints, including material gradation effects at both global and local levels. For instance, constraints associated with pattern repetition are applied by considering material gradation either on the global structure or locally over the specific pattern. By means of pattern repetition, we recover previous results in the literature which were obtained using homogenization and optimization of cellular materials.

Local-global processing in early-onset schizophrenia: Evidence for an impairment in shifting the spatial scale of attention

In this study we report the results of two experiments on visual attention conducted with patients with early-onset schizophrenia. These experiments investigated the effect of irrelevant spatial-scale information upon the processing of relevant spatial-scale information, and the ability to shift the spatial scale of attention, across consecutive trials, between different levels of the hierarchical stimulus. Twelve patients with early-onset schizophrenia and matched controls performed local-global tasks under: (1) directed attention conditions with a consistency manipulation and (2) divided-attention conditions. In the directed-attention paradigm, the early-onset patients exhibited the normal patterns of global advantage and interference, and were not unduly affected by the consistency manipulation. Under divided-attention conditions, however, the early-onset patients exhibited a local-processing deficit. The source of this local processing deficit lay in the prolonged reaction time to local targets, when these had been preceded by a global target, but not when preceded by a local target. These findings suggest an impaired ability to shift the spatial scale of attention from a global to a local spatial scale in early-onset schizophrenia. (C) 2003 Elsevier Science (USA). All rights reserved.