Previous Models Research Articles

Joint Entity and Relation Extraction With Set Prediction Networks.

Joint entity and relation extraction is an important task in natural language processing, which aims to extract all relational triples mentioned in a given sentence. In essence, the relational triples mentioned in a sentence are in the form of a set, which has no intrinsic order between elements and exhibits the permutation invariant feature. However, previous seq2seq-based models require sorting the set of relational triples into a sequence beforehand with some heuristic global rules, which destroys the natural set structure. In order to break this bottleneck, we treat joint entity and relation extraction as a direct set prediction problem, so that the extraction model is not burdened with predicting the order of multiple triples. To solve this set prediction problem, we propose networks featured by transformers with non-autoregressive parallel decoding. In contrast to autoregressive approaches that generate triples one by one in a specific order, the proposed networks are able to directly output the final set of relational triples in one shot. Furthermore, we also design a set-based loss that forces unique predictions through bipartite matching. Compared with cross-entropy loss that highly penalizes small shifts in triple order, the proposed bipartite matching loss is invariant to any permutation of predictions; thus, it can provide the proposed networks with a more accurate training signal by ignoring triple order and focusing on relation types and entities. Various experiments on two benchmark datasets demonstrate that our proposed model significantly outperforms the current state-of-the-art (SoTA) models. Training code and trained models are now publicly available at https://github.com/DianboWork/SPN4RE.

Recommending Target Actions Outside Sessions in the Data-poor Insurance Domain

Providing personalized recommendations for insurance products is particularly challenging due to the intrinsic and distinctive features of the insurance domain. First, unlike more traditional domains like retail, movie and so on, a large amount of user feedback is not available and the item catalog is smaller. Second, due to the higher complexity of products, the majority of users still prefer to complete their purchases over the phone instead of online. We present different recommender models to address such data scarcity in the insurance domain. We use recurrent neural networks with three different types of loss functions and architectures (cross-entropy, censored Weibull, and attention). Our models cope with data scarcity by learning from multiple sessions and different types of user actions. Moreover, differently from previous session-based models, our models learn to predict a target action that does not happen within the session. Our models outperform state-of-the-art baselines on a real-world insurance dataset, with ca. 44K users, 16 items, 54K purchases, and 117K sessions. Moreover, combining our models with demographic data boosts the performance. Analysis shows that considering multiple sessions and several types of actions are both beneficial for the models, and that our models are not unfair with respect to age, gender, and income.

An Explainable Multichannel Model for COVID-19 Time Series Prediction

Introduction: The COVID-19 pandemic has affected every country and changed people's lives. Accurate prediction of COVID-19 trends can help prevent the further spread of the outbreak. However, the changing environment affects the COVID-19 prediction performance, and previous models are limited in practical applications. Methods: An explainable multichannel deep learning model with spatial, temporal and environmental channels for time series prediction, STE-COVIDNet, was proposed. The time series data of COVID-19 infection, weather, in-state population mobility, and vaccination were collected from May, 2020, to October, 2021, in the USA. In the environmental channel of STE-COVIDNet, an attention mechanism was applied to extract significant environmental factors related to the spread of COVID-19. In addition, the attention weights of these factors were analyzed with the actual situation. Results: STE-COVIDNet was found to be superior to other advanced prediction models of COVID-19 infection cases. The analysis results of attention weight were reported to be consistent with existing studies and reports. It was found that the same environmental factors that influence the spread of COVID-19 can vary across time and region, which explains why findings of previous studies on the relationship between the environment and COVID-19 vary by region and time. Conclusion: STE-COVIDNet is an explainable model that can adapt to environmental changes and thus improve predictive performance.

GraphCL-DTA: A Graph Contrastive Learning With Molecular Semantics for Drug-Target Binding Affinity Prediction.

Drug-target binding affinity prediction plays an important role in the early stages of drug discovery, which can infer the strength of interactions between new drugs and new targets. However, the performance of previous computational models is limited by the following drawbacks. The learning of drug representation relies only on supervised data without considering the information in the molecular graph itself. Moreover, most previous studies tended to design complicated representation learning modules, while uniformity used to measure representation quality is ignored. In this study, we propose GraphCL-DTA, a graph contrastive learning with molecular semantics for drug-target binding affinity prediction. This graph contrastive learning framework replaces the dropout-based data augmentation strategy by performing data augmentation in the embedding space, thereby better preserving the semantic information of the molecular graph. A more essential and effective drug representation can be learned through this graph contrastive framework without additional supervised data. Next, we design a new loss function that can be directly used to adjust the uniformity of drug and target representations. By directly optimizing the uniformity of representations, the representation quality of drugs and targets can be improved. The effectiveness of the above innovative elements is verified on two real datasets, KIBA and Davis. Compared with the GraphDTA model, the relative improvement of the GraphCL-DTA model on the two datasets is 2.7% and 4.5%. The graph contrastive learning framework and uniformity function in the GraphCL-DTA model can be embedded into other computational models as independent modules to improve their generalization capability.

The gated cascade diffusion model: An integrated theory of decision making, motor preparation, and motor execution.

This article introduces an integrated and biologically inspired theory of decision making, motor preparation, and motor execution. The theory is formalized as an extension of the diffusion model, in which diffusive accumulated evidence from the decision-making process is continuously conveyed to motor areas of the brain that prepare the response, where it is smoothed by a mechanism that approximates a Kalman-Bucy filter. The resulting motor preparation variable is gated prior to reaching agonist muscles until it exceeds a particular level of activation. We tested this gated cascade diffusion model by continuously probing the electrical activity of the response agonists through electromyography in four choice tasks that span a variety of domains in cognitive sciences, namely motion perception, numerical cognition, recognition memory, and lexical knowledge. The model provided a good quantitative account of behavioral and electromyographic data and systematically outperformed previous models. This work represents an advance in the integration of processes involved in simple decisions and sheds new light on the interplay between decision and motor systems. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

KepSalinst: Using Peripheral Points to Delineate Salient Instances.

Salient instance segmentation (SIS) is an emerging field that evolves from salient object detection (SOD), aiming at identifying individual salient instances using segmentation maps. Inspired by the success of dynamic convolutions in segmentation tasks, this article introduces a keypoints-based SIS network (KepSalinst). It employs multiple keypoints, that is, the center and several peripheral points of an instance, as effective geometrical guidance for dynamic convolutions. The features at peripheral points can help roughly delineate the spatial extent of the instance and complement the information inside the central features. To fully exploit the complementary components within these features, we design a differentiated patterns fusion (DPF) module. This ensures that the resulting dynamic convolutional filters formed by these features are sufficiently comprehensive for precise segmentation. Furthermore, we introduce a high-level semantic guided saliency (HSGS) module. This module enhances the perception of saliency by predicting a map for the input image to estimate a saliency score for each segmented instance. On four SIS datasets (ILSO, SOC, SIS10K, and COME15K), our KepSalinst outperforms all previous models qualitatively and quantitatively.

Observational evidence to support a dense ambient medium shaping the jet in 3C 84

Highly collimated relativistic jets are a defining feature of certain active galactic nuclei (AGN), yet their formation mechanism remains elusive. Previous observations and theoretical models have proposed that the ambient medium surrounding the jets could exert pressure, playing a crucial role in shaping the jets. However, a direct observational confirmation of such a medium has been lacking. In this study, we present very long baseline interferometric (VLBI) observations of 3C 84 (NGC 1275), located at the center of the Perseus Cluster. Through monitoring observations with the Very Long Baseline Array (VLBA) at 43 GHz, a jet knot was detected to have been ejected from the sub-parsec scale core in the late 2010s. Intriguingly, this knot propagated in a direction significantly offset from the parsec-scale jet direction. To delve deeper into the matter, we employed follow-up VLBA 43 GHz observations, tracing the knot’s trajectory until the end of 2022. We discovered that the knot abruptly changed its trajectory in the early 2020s, realigning itself with the parsec-scale jet direction. Additionally, we present results from an observation of 3C 84 with the Global VLBI Alliance (GVA) at 22 GHz, conducted near the monitoring period. By jointly analyzing the GVA 22 GHz image with a VLBA 43 GHz image observed about one week apart, we generated a spectral index map, revealing an inverted spectrum region near the edge of the jet where the knot experienced deflection. These findings suggest the presence of a dense, cold ambient medium characterized by an electron density exceeding ∼105 cm−3, which guides the jet’s propagation on parsec scales and significantly contributes to the overall shaping of the jet.

Co-Guiding for Multi-Intent Spoken Language Understanding.

Recent graph-based models for multi-intent SLU have obtained promising results through modeling the guidance from the prediction of intents to the decoding of slot filling. However, existing methods (1) only model the unidirectional guidance from intent to slot, while there are bidirectional inter-correlations between intent and slot; (2) adopt homogeneous graphs to model the interactions between the slot semantics nodes and intent label nodes, which limit the performance. In this paper, we propose a novel model termed Co-guiding Net, which implements a two-stage framework achieving the mutual guidances between the two tasks. In the first stage, the initial estimated labels of both tasks are produced, and then they are leveraged in the second stage to model the mutual guidances. Specifically, we propose two heterogeneous graph attention networks working on the proposed two heterogeneous semantics-label graphs, which effectively represent the relations among the semantics nodes and label nodes. Besides, we further propose Co-guiding-SCL Net, which exploits the single-task and dual-task semantics contrastive relations. For the first stage, we propose single-task supervised contrastive learning, and for the second stage, we propose co-guiding supervised contrastive learning, which considers the two tasks' mutual guidances in the contrastive learning procedure. Experiment results on multi-intent SLU show that our model outperforms existing models by a large margin, obtaining a relative improvement of 21.3% over the previous best model on MixATIS dataset in overall accuracy. We also evaluate our model on the zero-shot cross-lingual scenario and the results show that our model can relatively improve the state-of-the-art model by 33.5% on average in terms of overall accuracy for the total 9 languages.

Efficient disease identification using symptom-based ensemble models and bayes-search optimization

Symptom-based disease identification is crucial to the diagnosis of the disease at the early stage. Usage of traditional stacking and blending models i.e., with default values of the models cannot handle the multi-classification data properly. Some of the existing researchers also implemented tuning with the help of a grid search approach but it consumes more time because it checks all the possible combinations. Suppose if the model has n estimators with k values it needs to check (n*k)! elements combination, this makes the learning time high. The proposed model chooses the estimators to train the model with in a considerable amount of time using an advanced tuning technique known as “Bayes-Search” on an ensemble random forest and traditional, support vector machine. The advantage of this model is its capability to store the best evaluation metrics from the previous model and utilise them to store the new model. This model chooses the values of the estimator based on the probability of selection, which reduces the elements in search space i.e., (< (n-k)!). The proposed model defines the objective function with a minimum error rate and predicts the error rate with the selected estimators for different distributions. The model depending on the predicted value decides whether to store the value or to return the value to the optimizer. The Bayes search optimization has achieved +9.21% accuracy than the grid search approach. Among the two approaches random forest has achieved good accuracy and less loss using Bayes search with cross-validation.

A 3-D Convolution-Based Burn-Through Point Multistep Prediction Model for Sintering Process

Predicting burn-through point (BTP) in advance is a quite critical task for the sintering process. However, sintering is a complex physicochemical reaction process, and the strong spatial-temporal correlations of data make the multi-step prediction task very challenging. The previous BTP multi-step prediction model only extracts spatial features in the high-level layers, leaving the spatial features in the low-level layers not learned. Specifically, the previous model only considers the relationships between the process variables and BTP, ignoring the spatial coupling relations among process variables. Further, the existing loss function is mainly based on Euclidean distance, which cannot learn dynamic information of multi-step prediction sequence. To tackle these problems, in this study, we propose a 3D convolution-based BTP multi-step prediction model (CBMP) to simultaneously capture spatial-temporal features. First, the 3D convolution is employed to capture the spatial-temporal features from low-level to high-level layers at the same time. Secondly, a spatial-temporal recalibration block is proposed to further refine the extracted features to increase the contributions of informative features and suppress the less useful ones. Finally, we design a time-aware multi-step prediction loss function to dynamically weigh the similarity between the actual sequence and the predicted sequence. The experimental results on two real-world BTP datasets demonstrate the effectiveness and feasibility of the proposed model on the BTP multi-step prediction task.

Learning to Synthesize Compatible Fashion Items Using Semantic Alignment and Collocation Classification: An Outfit Generation Framework.

The field of fashion compatibility learning has attracted great attention from both the academic and industrial communities in recent years. Many studies have been carried out for fashion compatibility prediction, collocated outfit recommendation, artificial intelligence (AI)-enabled compatible fashion design, and related topics. In particular, AI-enabled compatible fashion design can be used to synthesize compatible fashion items or outfits to improve the design experience for designers or the efficacy of recommendations for customers. However, previous generative models for collocated fashion synthesis have generally focused on the image-to-image translation between fashion items of upper and lower clothing. In this article, we propose a novel outfit generation framework, i.e., OutfitGAN, with the aim of synthesizing a set of complementary items to compose an entire outfit, given one extant fashion item and reference masks of target synthesized items. OutfitGAN includes a semantic alignment module (SAM), which is responsible for characterizing the mapping correspondence between the existing fashion items and the synthesized ones, to improve the quality of the synthesized images, and a collocation classification module (CCM), which is used to improve the compatibility of a synthesized outfit. To evaluate the performance of our proposed models, we built a large-scale dataset consisting of 20 000 fashion outfits. Extensive experimental results on this dataset show that our OutfitGAN can synthesize photo-realistic outfits and outperform the state-of-the-art methods in terms of similarity, authenticity, and compatibility measurements.

Robust Polar Amplification in Ice-Free Climates Relies on Ocean Heat Transport and Cloud Radiative Effects

Abstract A fundamental divide exists between previous studies that conclude that polar amplification does not occur without sea ice and studies that find that polar amplification is an inherent feature of the atmosphere independent of sea ice. We hypothesize that a representation of climatological ocean heat transport is key for simulating polar amplification in ice-free climates. To investigate this, we run a suite of targeted experiments in the slab ocean aquaplanet configuration of CESM2-CAM6 with different profiles of prescribed ocean heat transport, which are invariant under CO2 quadrupling. In simulations without climatological ocean heat transport, polar amplification does not occur. In contrast, in simulations with climatological ocean heat transport, robust polar amplification occurs in all seasons. What is causing this dependence of polar amplification on ocean heat transport? Energy-balance model theory is incapable of explaining our results and in fact would predict that introducing ocean heat transport leads to less polar amplification. We instead demonstrate that shortwave cloud radiative feedbacks can explain the divergent polar climate responses simulated by CESM2-CAM6. Targeted cloud locking experiments in the zero ocean heat transport simulations are able to reproduce the polar amplification of the climatological ocean heat transport simulations, solely by prescribing high-latitude cloud radiative feedbacks. We conclude that polar amplification in ice-free climates is underpinned by ocean–atmosphere coupling, through a less negative high latitude shortwave cloud radiative feedback that facilitates enhanced polar warming. In addition to reconciling previous disparities, these results have important implications for interpreting past equable climates and climate projections under high-emissions scenarios. Significance Statement Polar amplification is a robust feature of climate change in the modern-day climate. However, previous climate modeling studies fundamentally do not agree on whether polar amplification occurs in ice-free climates. In this study, we find in a state-of-the-art climate model that, if ocean heat transport is neglected, the response to an increase in CO2 is not polar amplified, whereas robust polar amplification occurs if ocean heat transport is included. Using targeted model experiments, we diagnose cloud radiative effects as the driver of this divergent behavior. We conclude that polar amplification is a robust feature of the atmosphere–ocean system. Our results have important implications for interpreting past warm climates and future projections under high-emissions scenarios.

Magnetically driven winds from accretion disks in post-asymptotic giant branch binaries

Context. Jets are commonly detected in post-asymptotic giant branch (post-AGB) binaries and originate from an accretion process onto the companion of the post-AGB primary. These jets are revealed by high-resolution spectral time series. Aims. This paper is part of a series. In this work, we move away from our previous parametric modelling and include a self-similar wind model that allows the physical properties of post-AGB binaries to be characterised. This model describes magnetically driven jets from a thin accretion disk threaded by a large-scale, near equipartition vertical field. Methods. We expanded our methodology in order to simulate the high-resolution dynamic spectra coming from the obscuration of the primary by the jets launched by the companion. We present the framework to exploit the self-similar jet models for post-AGB binaries. We performed a parameter study to investigate the impact of different parameters (inclination, accretion rate, inner and outer launching radius) on the synthetic spectra. Results. We successfully included the physical jet models into our framework. The synthetic spectra have a very similar orbital phase coverage and absorption strengths as the observational data. The magnetohydrodynamic (MHD) jet models provide a good representation of the actual jet creation process in these evolved binaries. Challenges remain, however, as the needed high-accretion rate would induce accretion disks that are too hot in comparison to the data. Moreover, the rotational signature of the models is not detected in the observations. In future research, we will explore models with a higher disk ejection efficiency and even lower magnetisation in order to solve some of the remaining discrepancies between the observed and synthetic dynamic spectra.

A lightweight security model for ensuring patient privacy and confidentiality in telehealth applications

Telehealth, a cornerstone of Healthcare 4.0 has widespread adoption owing to its convenience and efficiency. However, the secure transmission of patient records over the Internet and Cloud services remains a paramount concern. This study endeavors to design an enhanced security model that guarantees the privacy and confidentiality of patient electronic health records in telehealth applications. The proposed model employs a lightweight encryption algorithm to generate an access token that strictly regulates data access exclusively to the patient's department. Leveraging the patient's unique ID for encryption and decryption processes, the model ensures the secure handling of sensitive information. Robust data privacy measures are implemented through a random key generator utilizing the patient's ECG signal for secure key generation. Test results underscore the efficacy of the proposed security model in limiting access to patient-related data within the designated department. Furthermore, the model employs an efficient encryption algorithm, resulting in a 0.2% increase in generation rate compared to previous models, while reducing CPU usage by 23.16%. These enhancements significantly bolster the reliability and efficiency of the security model, providing robust safeguards for patient electronic health records' privacy and confidentiality during transmission across diverse telehealth applications. This study makes a substantial contribution to healthcare by addressing critical concerns surrounding patient privacy and confidentiality. The designed security model stands poised for seamless integration into various telehealth applications, offering a steadfast solution for data privacy and confidentiality concerns.

Enhancing cladding mechanical modelling during DBA/LOCA accidents with FRAPTRAN: The TUmech one-dimensional model

The analysis of the cladding mechanical performance is essential for predicting rod failure during the heat-up phase of a loss-of-coolant accident (LOCA). Previous modelling and simulation exercises have shown that there is room for improvement in the accuracy of fuel performance simulations under such conditions. In this work, TUmech, a simplified standalone version of the mechanical model in TRANSURANUS, has been developed and coupled with FRAPTRAN-2.0. After highlighting the most significant differences between both approaches, particularly in their mechanical treatment and material property correlations, simulations have been performed for IFA-650.2 (fresh fuel) and IFA-650.10 (high burnup fuel) of the Halden LOCA test series. The results obtained from both FRAPTRAN versions have been compared with experimental data and benchmarked against TRANSURANUS estimates, focusing on rod inner gas pressure, post-test cladding outer diameter, and burst time. In both scenarios, FRAPTRAN-TUmech has proven to enhance the accuracy in predicting these three variables compared to the FRAPTRAN default version. However, further validation is required before drawing definitive conclusions based on the encouraging results.

Exploring ChatGPT for next-generation information retrieval: Opportunities and challenges

The rapid advancement of artificial intelligence (AI) has spotlighted ChatGPT as a key technology in the realm of information retrieval (IR). Unlike its predecessors, it offers notable advantages that have captured the interest of both industry and academia. While some consider ChatGPT to be a revolutionary innovation, others believe its success stems from smart product and market strategy integration. The advent of ChatGPT and GPT-4 has ushered in a new era of Generative AI, producing content that diverges from training examples, and surpassing the capabilities of OpenAI’s previous GPT-3 model. In contrast to the established supervised learning approach in IR tasks, ChatGPT challenges traditional paradigms, introducing fresh challenges and opportunities in text quality assurance, model bias, and efficiency. This paper aims to explore the influence of ChatGPT on IR tasks, providing insights into its potential future trajectory.

DocFormerv2: Local Features for Document Understanding

We propose DocFormerv2, a multi-modal transformer for Visual Document Understanding (VDU). The VDU domain entails understanding documents (beyond mere OCR predictions) e.g., extracting information from a form, VQA for documents and other tasks. VDU is challenging as it needs a model to make sense of multiple modalities (visual, language and spatial) to make a prediction. Our approach, termed DocFormerv2 is an encoder-decoder transformer which takes as input - vision, language and spatial features. DocFormerv2 is pre-trained with unsupervised tasks employed asymmetrically i.e., two novel document tasks on encoder and one on the auto-regressive decoder. The unsupervised tasks have been carefully designed to ensure that the pre-training encourages local-feature alignment between multiple modalities. DocFormerv2 when evaluated on nine challenging datasets shows state-of-the-art performance on all over strong baselines - On TabFact (+4.3%), InfoVQA (+1.4%), FUNSD (+1.0%). Furthermore, to show generalization capabilities, on three VQA tasks involving scene-text, DocFormerv2 outperforms previous comparably-sized models and even does better than much larger models (such as GIT2, PaLI and Flamingo) on these tasks. Extensive ablations show that due to its novel pre-training tasks, DocFormerv2 understands multiple modalities better than prior-art in VDU.

Semantic Segmentation in Multiple Adverse Weather Conditions with Domain Knowledge Retention

Semantic segmentation's performance is often compromised when applied to unlabeled adverse weather conditions. Unsupervised domain adaptation is a potential approach to enhancing the model's adaptability and robustness to adverse weather. However, existing methods encounter difficulties when sequentially adapting the model to multiple unlabeled adverse weather conditions. They struggle to acquire new knowledge while also retaining previously learned knowledge. To address these problems, we propose a semantic segmentation method for multiple adverse weather conditions that incorporates adaptive knowledge acquisition, pseudo-label blending, and weather composition replay. Our adaptive knowledge acquisition enables the model to avoid learning from extreme images that could potentially cause the model to forget. In our approach of blending pseudo-labels, we not only utilize the current model but also integrate the previously learned model into the ongoing learning process. This collaboration between the current teacher and the previous model enhances the robustness of the pseudo-labels for the current target. Our weather composition replay mechanism allows the model to continuously refine its previously learned weather information while simultaneously learning from the new target domain. Our method consistently outperforms the state-of-the-art methods, and obtains the best performance with averaged mIoU (%) of 65.7 and the lowest forgetting (%) of 3.6 against 60.1 and 11.3, on the ACDC datsets for a four-target continual multi-target domain adaptation.

Hawkes-Enhanced Spatial-Temporal Hypergraph Contrastive Learning Based on Criminal Correlations

Crime prediction is a crucial yet challenging task within urban computing, which benefits public safety and resource optimization. Over the years, various models have been proposed, and spatial-temporal hypergraph learning models have recently shown outstanding performances. However, three correlations underlying crime are ignored, thus hindering the performance of previous models. Specifically, there are two spatial correlations and one temporal correlation, i.e., (1) co-occurrence of different types of crimes (type spatial correlation), (2) the closer to the crime center, the more dangerous it is around the neighborhood area (neighbor spatial correlation), and (3) the closer between two timestamps, the more relevant events are (hawkes temporal correlation). To this end, we propose Hawkes-enhanced Spatial-Temporal Hypergraph Contrastive Learning framework (HCL), which mines the aforementioned correlations via two specific strategies. Concretely, contrastive learning strategies are designed for two spatial correlations, and hawkes process modeling is adopted for temporal correlations. Extensive experiments demonstrate the promising capacities of HCL from four aspects, i.e., superiority, transferability, effectiveness, and sensitivity.

A quantitative genetic model of background selection in humans.

Across the human genome, there are large-scale fluctuations in genetic diversity caused by the indirect effects of selection. This "linked selection signal" reflects the impact of selection according to the physical placement of functional regions and recombination rates along chromosomes. Previous work has shown that purifying selection acting against the steady influx of new deleterious mutations at functional portions of the genome shapes patterns of genomic variation. To date, statistical efforts to estimate purifying selection parameters from linked selection models have relied on classic Background Selection theory, which is only applicable when new mutations are so deleterious that they cannot fix in the population. Here, we develop a statistical method based on a quantitative genetics view of linked selection, that models how polygenic additive fitness variance distributed along the genome increases the rate of stochastic allele frequency change. By jointly predicting the equilibrium fitness variance and substitution rate due to both strong and weakly deleterious mutations, we estimate the distribution of fitness effects (DFE) and mutation rate across three geographically distinct human samples. While our model can accommodate weaker selection, we find evidence of strong selection operating similarly across all human samples. Although our quantitative genetic model of linked selection fits better than previous models, substitution rates of the most constrained sites disagree with observed divergence levels. We find that a model incorporating selective interference better predicts observed divergence in conserved regions, but overall our results suggest uncertainty remains about the processes generating fitness variation in humans.