In the world of the Internet and World Wide Web, which offers a tremendous amount of information, an increasing emphasis is being given to searching services and functionality. Currently, a majority of web portals offer their searching utilities, be it better or worse. These can search for the content within the sites, mainly text the textual content of documents. In this paper a novel similarity measure called SMDR (Similarity Measure for Documents Retrieval) is proposed to help retrieve more similar documents from the repository thus contributing considerably to the effectiveness of Web Information Retrieval (WIR) process. Bio-inspired PSO methodology is used with the intent to reduce the response time of the system and optimizes WIR process, hence contributes to the efficiency of the system. This paper also demonstrates a comparative study of the proposed system with the existing method in terms of accuracy, sensitivity, F-measure and specificity. Finally, extensive experiments are conducted on CACM collections. Better precision-recall rates are achieved than the existing system. Experimental results demonstrate the effectiveness and efficiency of the proposed system.

We propose a selective learning method using meta-learning and deep reinforcement learning for medical image interpretation in the setting of limited labeling resources. Our method, MedSelect, consists of a trainable deep learning selector that uses image embeddings obtained from contrastive pretraining for determining which images to label, and a non-parametric selector that uses cosine similarity to classify unseen images. We demonstrate that MedSelect learns an effective selection strategy outperforming baseline selection strategies across seen and unseen medical conditions for chest X-ray interpretation. We also perform an analysis of the selections performed by MedSelect comparing the distribution of latent embeddings and clinical features, and find significant differences compared to the strongest performing baseline. We believe that our method may be broadly applicable across medical imaging settings where labels are expensive to acquire.

Air pollution from fine particulate matter (PM2.5) is increasingly driving the global burden of disease ([ 1 ][1]), and diesel-powered vehicles are substantial contributors. Recognizing the public health impacts of diesel PM2.5 (DPM) ([ 2 ][2]), many countries have reduced emissions of DPM from both on- and off-road mobile sources over the past three decades. The previous US federal administration, however, changed course by eliminating or weakening policies and standards that govern these emissions. In contrast, the State of California has continued to reduce mobile-source DPM emissions using the state's long-standing authority under the Clean Air Act (CAA) to regulate air pollution more stringently than the federal government. Our analysis of mobile-source DPM emissions suggests that many California sector-based policies have been highly effective relative to the rest of the US. To improve health in communities disproportionately affected by these emissions, we point to opportunities to further reduce DPM emissions in California, in the US more broadly, and in parts of the world where countries have less aggressive vehicle emissions policies than the US ([ 3 ][3]). The US has targeted emissions of nitrogen oxides (NO x ) and DPM from diesel trucks and buses, railway locomotives, marine vessels, and off-road engines used in construction and agriculture through successively tighter emissions standards phased in since 1994 (table S1). These standards require low- and ultralow-sulfur diesel fuels (LSDF and ULSDF), establish emissions limits, and institute systems for portable emissions measurement and onboard diagnostics (table S1). The US Environmental Protection Agency (EPA) estimated that full implementation of Obama-era US emissions standards by 2030 would prevent some 12,000 premature deaths annually ([ 4 ][4]). Despite this, EPA leadership disbanded the PM review panel ahead of the scheduled 2020 update of federal PM standards; it also rolled back, or attempted to roll back, 85 federal air pollution policies ([ 5 ][5]) and moved to restrict the ability of states to set more stringent emissions standards ([ 6 ][6]). California, whose economy would rank fifth largest in the world if it were a sovereign nation, hosts the country's two largest ports and moves 60% of its container cargo (see supplementary materials). With the associated truck and rail traffic, California stands out as the largest emitter of DPM in the country. At the same time, California has also led the nation with the largest overall reduction in metric tons of DPM emissions from mobile sources. Over the past three decades, California's policies have systematically targeted high-emitting sectors, reducing mobile-source DPM emissions by, for example, substituting electric for diesel power where feasible, tightening emissions limits for new and existing diesel engines, and requiring ULSDF, which emits substantially less PM2.5 than higher-sulfur fuels upon combustion and can be combined with particle filters to further reduce emissions. To understand the impact of California's portfolio of policies, we used DPM emissions data from the EPA National Emissions Inventory (NEI), which assembles a comprehensive estimate of air pollution emissions using data reported by states, combined with modeled and measured inputs. We compared mobile-source DPM emissions in California versus the rest of the US for the period 1990 to 2014, the earliest and most recent year for which consistent NEI data are available ([ 7 ][7]). During that time, California reduced overall mobile-source DPM emissions by 78% while the rest of the US saw only a 51% reduction. These reductions came despite a concurrent steady rise in diesel fuel consumption: 20% in California and 28% in the rest of the US (data S1). Emissions reductions from heavy-duty diesel vehicles (HDDVs)—commercial trucks and buses—caused most of this decline, accounting for 67% of DPM emissions reductions in California and 57% in the rest of the US. Although the federal phase-in of ULSDF, off-road emissions standards, and the Heavy-Duty Engine and Vehicle Rule has reduced HDDV emissions across the US, California's reductions from HDDVs have been steeper and contribute even more to the overall reductions than would be predicted from the sector's size. Analyses of DPM emissions over time and the relative contributions made by each sector point to the effectiveness of California's policies that require diesel engine retrofits (adding emissions controls to existing HDDVs) and early replacement of older engines with newer, cleaner engines. Our analysis identifies three distinct phases in mobile-source DPM emissions between 1990 and 2014. Emissions fell overall from 1990 to 2001 in California and from 1990 to 2005 in the rest of the country. Reduced emissions from HDDVs contributed the largest share of the overall drop (see the figure and data S1). These changes are attributable to the introduction of LSDF nationwide, and to California's new requirements for vehicle inspections (table S2). Then, from 2001 to 2005 in California and from 2005 to 2008 in the rest of the country, emissions rose during an economic boom, driven primarily by increasing emissions from HDDVs and marine sources. Finally, overall DPM emissions once again fell, beginning in California in 2005 and in the rest of the US in 2008. The recession played a role in the early part of this drop ([ 8 ][8]), but emissions reductions continued through 2014 despite the economic recovery and the corresponding upturn in diesel use. During this final phase, California's 67% drop in DPM emissions outpaced the 40% reduction seen in the rest of the country (see the figure and data S1). Our analysis of individual sectors and each state's HDDV emissions suggests that California policies specifically targeting emissions from HDDVs and marine sources drove this decline. The later phases of California's emissions reductions correspond to the implementation of two overarching plans by the California Air Resources Board (CARB): the Diesel Risk Reduction Plan and the Emission Reduction Plan for Ports and Goods Movement (Goods Movement Plan), both of which encompassed multiple policies governing emissions from trucks and buses, ports, and off-road engines (table S2). Key policies targeting on-road HDDVs took effect in 2006 and 2007, further lowering the sulfur content of diesel fuel to 15 ppm (table S2) and tightening DPM emissions standards by 90% for new HDDVs (table S2). Beginning in 2010, with a rolling compliance period starting in 2015, all on-road HDDVs that operate in California were required to either retrofit existing engines with particle filters or replace engines older than the 2007 model year (table S2). By comparison, federal policies do not require retrofit or replacement of old diesel engines to meet emission standards, and HDDV engines typically operate for almost two decades, or about a million miles, before retirement. Our state-level analysis shows that by 2014 California HDDVs were emitting 139 metric tons of DPM for every billion vehicle-miles traveled (VMT), far less than the next-closest state (Oklahoma, 250 metric tons DPM per billion VMT) and the average in the rest of the country (345 metric tons DPM per billion VMT) (data S1). Although HDDVs remain California's largest source of DPM emissions, regulatory actions by CARB (over and above federal standards) have reduced HDDV emissions by 85% since 1990. If California's HDDV sector had followed the trajectory of other US states and DC, HDDV emissions in the state would have dropped only 58% (95% confidence interval, 52 to 64%) in that period (data S1). Also notable is the impact of two key CARB policies targeting marine sources. The 2007 At-Berth rule requires that oceangoing vessels switch to electric shore power while in port or use alternative control technologies to reduce emissions by an equivalent amount (table S2). The Cleaner Ocean Vessel fuel policy, finalized in 2008, requires that ships within 24 nautical miles of California's shoreline replace heavy fuel oil in their main engines with lower-sulfur fuels (table S2). Between 2008 and 2014, marine DPM emissions in the state dropped 51% overall (see the figure and data S1), and by 2018 emissions measured at the Port of Los Angeles had declined by 37% (fig. S3, A and B, and data S1). ![Figure][9] California versus the rest of the United States: Mobile-source DPM emissions declined differently Mobile-source diesel PM2.5 (DPM) emissions by sector in California versus the rest of the US from 1990 to 2014. HDDV, heavy-duty diesel vehicle; LDDV, light-duty diesel vehicle. All percentage changes reflect values relative to 1990 values. GRAPHIC: N. CARY/ SCIENCE By contrast, California has struggled to target diesel emissions from agriculture (table S2). The sector is responsible for up to 18% of the state's total DPM emissions from mobile sources, but it accounted for less than 1% of the total emissions reductions in California between 1990 and 2014. Although these figures do not reflect gains from voluntary tractor engine replacements that are reported differently, opportunities remain to reduce off-road farm emissions in the nation's leading agricultural state. Voluntary programs have further reduced DPM emissions beyond California's regulatory requirements. Incentives to bring engines and equipment to a standard cleaner than required by law are estimated to have reduced DPM emissions by more than 6000 metric tons since 2001 (table S2). A program established in 2006 has provided $1 billion in grants to update trucks, locomotives, and ships at berth, eliminating an estimated 2200 metric tons of DPM emissions (table S2). Like other policies targeting emissions along goods-movement corridors, this program particularly benefits neighboring communities, which tend to be lower-income communities of color (table S4). Taken together, CARB's policies reduced emissions to the extent that by 2014 California was emitting less than half the DPM that would be expected had the state followed the same trajectory as the rest of the US (fig. S2 and data S1). Correspondingly, we estimate that more than twice as many Californians would have died from DPM-attributable cardiopulmonary disease in 2014 alone if the state had not so markedly reduced emissions (data S1). The impact of targeted emissions regulation is also evident nationally, but it has come later and never as meaningfully as in California. Farming and construction emissions fell following the 2007 EPA Heavy Duty Engine and Vehicle Rule and the 2008–2015 phase-in of Tier 4 standards targeting off-road emissions from farm and construction equipment (table S1). Federal requirements for LSDF in the 1990s and ULSDF beginning in 2006 reduced HDDV emissions from both nonroad and on-road sources (table S1). In the marine sector, US coastal areas caught up to California's fuel standards in 2012 when ULSDF was required for smaller marine engines (table S1) and in 2015 for the largest vessels when requirements for lower-sulfur marine diesel came into effect in the North American Emissions Control Area established by the International Maritime Organization (table S1). By contrast, California has taken not only earlier action on marine emissions but also aggressive steps to target emissions from the many engines that pollute the air near ports, including marine auxiliary engines, short-haul trucks, cargo-handling cranes, and yard trucks (table S2). Individual states that have reduced HDDV emissions more than the national average are more likely to have adopted California's standards, as permitted under the CAA (table S5 and data S1), and the rest of the US could do the same. Coordination across states and between state and federal agencies means that methodological differences in data collection are unlikely to account for the observed differences in DPM emissions between California and the rest of the US (see supplementary materials). But how do we know that emission inventories are accurate and, furthermore, that CARB policies are responsible for the observed reductions? Field studies measuring changes in concentrations of DPM serve to ground-truth emissions inventories and substantiate the link between policy interventions and observed outcomes (table S4). For example, following the suite of interventions under the 2006 Goods Movement Plan, California communities in close proximity to goods-movement corridors saw significantly greater air quality improvements relative to non–goods-movement corridors and control areas monitored during the same time period (table S4). These findings show specific, local impacts of regulations targeting high-emitting sectors, distinguishing those changes from secular trends in air pollution and demonstrating their potential to advance environmental justice. The 2007 CARB regulation requiring retrofit or replacement of older HDDV engines for short-haul “drayage trucks” that operate at ports and railyards corresponded to a 70% reduction in black carbon emissions (a DPM proxy) and a 75% reduction in PM mass specific to drayage trucks measured in and around the ports of Oakland and Los Angeles between 2009 and 2011 (table S4). These changes mirror the emissions reductions measured in laboratory testing of the low-sulfur fuels and retrofit technologies used to meet the drayage truck standards (table S3). Likewise, the 2009 CARB requirement for low-sulfur fuels in oceangoing vessel engines operating within 24 nautical miles of the California coastline was associated with a measured 64% drop in San Francisco Bay Area concentrations of vanadium, a marker for combustion of heavy fuel oil (table S4). Sampling conducted by aircraft flying in the exhaust plume of a container ship approaching the coast showed that the fuel switch, combined with a required speed reduction, dropped DPM emissions by 90% (table S4). That these changes all occurred in the setting of continued growth in California's population, gross state product, and diesel consumption (figs. S4 and S5) further supports the assertion that the observed reductions track to the policies targeting DPM emissions. Observed emissions reductions are further corroborated by epidemiological data that link specific CARB policies to regional reductions in children's exposure to particle pollution and show corresponding improvements in both lung function and development in children with and without asthma ([ 9 ][10]). Finally, comparing HDDV sector emissions in California to the rest of the country likely underestimates the actual impact of CARB policies, which apply not only to the nearly half-million trucks and buses registered in California but also to the same number of out-of-state HDDVs estimated to drive California's highways each year ([ 10 ][11]). This requirement reduces emissions outside of California as well, although those reductions are attributed to federal policy. In California, cleaner air has not come at the expense of the state's economy, which in recent years has grown at double the average national rate ([ 11 ][12]). CARB estimates that every dollar the state has spent controlling air pollution has generated $38 in benefits attributable to reduced air pollution–related illness, premature death, and lost productivity. California's overall economic gain from health benefits linked to air pollution reduction, including CARB rules and programs, is estimated to have exceeded $250 billion between 1973 and 2014 ([ 12 ][13]). The link between PM2.5 exposure and increased risk of hospitalization and death from COVID-19 ([ 13 ][14]) further underscores the public health importance of cleaner air, particularly for communities of color that are disproportionately affected by both. California could benefit from additional measures to reduce emissions from off-road sectors, such as construction and agriculture, which CARB has not tackled as aggressively ([ 14 ][15]). Indeed, the nation as a whole could reduce mobile-source DPM emissions by requiring ships at berth to use shore power, and by requiring replacement or retrofit of existing on-road and off-road HDDVs in advance of fleet turnover. Given the long service life of older, dirty diesel engines, the current federal policy of mandating engine upgrades only with vehicle turnover is simply too slow. As the US initiates new federal rule-making on the proposed Cleaner Trucks Initiative to reduce NO x emissions from HDDVs, industry and environmental groups are calling on EPA to address NO x and DPM emissions in tandem and to create consistent “50-state” standards ([ 15 ][16]). In doing so, the EPA should align with CARB rules. EPA should also remove federal preemption of state emissions limits for off-road engines used in construction and agriculture. Even absent more aggressive federal policy, states' authority to set and implement their own stricter emissions standards must be protected. [science.sciencemag.org/content/371/6536/1314/suppl/DC1][17] 1. [↵][18]GBD 2017 Risk Factor Collaborators, Lancet 392, 1923 (2018). [OpenUrl][19][CrossRef][20][PubMed][21] 2. [↵][22]California Air Resources Board, “Overview: Diesel Exhaust & Health”; [ww2.arb.ca.gov/resources/overview-diesel-exhaust-and-health][23]. 3. [↵][24]European Union Directorate-General for Internal Policies, Comparative Study on the Differences Between the EU and US Legislation on Emissions in the Automotive Sector (2016); [www.europarl.europa.eu/RegData/etudes/STUD/2016/587331/IPOL\_STU(2016)587331\_EN.pdf][25]. 4. [↵][26]DieselNet, “Emission Standards, United States”; [www.dieselnet.com/standards/us/index.php][27]. 5. [↵][28]1. J. M. Samet, 2. T. A. Burke , Annu. Rev. Public Health 41, 347 (2020). [OpenUrl][29] 6. [↵][30]1. C. Davenport , “Trump to Revoke California's Authority to Set Stricter Auto Emissions Rules.” New York Times, 17 September 2019; [www.nytimes.com/2019/09/17/climate/trump-california-emissions-waiver.html][31]. 7. [↵][32]US Environmental Protection Agency, “National Emissions Inventory (NEI)”; [www.epa.gov/air-emissions-inventories/national-emissions-inventory-nei][33]. 8. [↵][34]1. D. Q. Tong et al ., Atmos. Environ. 107, 70 (2015). [OpenUrl][35] 9. [↵][36]1. F. Gilliland et al ., “The Effects of Policy-Driven Air Quality Improvements on Children's Respiratory Health” (2017); [www.healtheffects.org/system/files/GillilandRR190.pdf][37]. 10. [↵][38]California Air Resources Board, “Staff Report: Initial Statement of Reasons for Proposed Rulemaking: Proposed Regulation for In-Use On-Road Diesel Vehicles” (2008); . 11. [↵][39]Next10, 2017 California Green Innovation Index (2017); [www.next10.org/publications/2017-gii][40]. 12. [↵][41]California Air Resources Board, Fifty Year Air Quality Trends and Health Benefits; [ww3.arb.ca.gov/board/books/2018/020818/18-1-2pres.pdf][42]. 13. [↵][43]1. X. Wu, 2. R. C. Nethery, 3. M. B. Sabath, 4. D. Braun, 5. F. Dominici , Air pollution and COVID-19 mortality in the United States: Strengths and limitations of an ecological regression analysis. Sci. Adv. 6, eabd4049 (2020). 10.1126/sciadv.abd4049pmid:33148655 [OpenUrl][44][FREE Full Text][45] 14. [↵][46]California's construction emissions declined markedly from 2008 to 2011. Although industry likely lowered emissions in anticipation of deadlines in the 2008 In-Use Off-Road Diesel-Fueled Fleet Regulation (table S2), the majority of the decline is likely attributable to CARB's 2011 construction inventory revision prompted by the regulated industry. In that year, the regulation was also amended to delay implementation by 4 years and to lower required emission reductions. 15. [↵][47]US Environmental Protection Agency, “Control of Air Pollution From New Motor Vehicles: Heavy-Duty Engine Standards” [proposed rule]; [www.federalregister.gov/documents/2020/01/21/2020-00542/control-of-air-pollution-from-new-motor-vehicles-heavy-duty-engine-standards#citation-4-p3307][48]. Acknowledgments: We thank K. Peterson (University of California, Berkeley) for data visualization; K. Karparos, C. Parmer, and B. Holmes-Gen (CARB) for manuscript review; M. Houyoux, J. Godfrey, and M. Aldrich (EPA) for assistance with NEI data; and J. Austin, R. Boyd, T. Brasil, J. Cao, M. Diaz, R. Furey, J. Herner, S. Huber, M. Komlenic, R. Krieger, T. Kuwayama, N. Lowery, N. Motallebi, S. Pournazeri, S. Yoon, S. Zelinka, and L. Zhou (CARB) for assistance with CARB regulations and data. This research was supported in part by California Breast Cancer Research Program grant 23QB-1881. J.B. serves as the Physician Member of CARB. A.A. is a former employee of CARB. [1]: #ref-1 [2]: #ref-2 [3]: #ref-3 [4]: #ref-4 [5]: #ref-5 [6]: #ref-6 [7]: #ref-7 [8]: #ref-8 [9]: pending:yes [10]: #ref-9 [11]: #ref-10 [12]: #ref-11 [13]: #ref-12 [14]: #ref-13 [15]: #ref-14 [16]: #ref-15 [17]: http://science.sciencemag.org/content/371/6536/1314/suppl/DC1 [18]: #xref-ref-1-1 "View reference 1 in text" [19]: {openurl}?query=rft.jtitle%253DLancet%26rft.volume%253D392%26rft.spage%253D1923%26rft_id%253Dinfo%253Adoi%252F10.1016%252FS0140-6736%252818%252932225-6%26rft_id%253Dinfo%253Apmid%252F30496105%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [20]: /lookup/external-ref?access_num=10.1016/S0140-6736(18)32225-6&link_type=DOI [21]: /lookup/external-ref?access_num=30496105&link_type=MED&atom=%2Fsci%2F371%2F6536%2F1314.atom [22]: #xref-ref-2-1 "View reference 2 in text" [23]: http://ww2.arb.ca.gov/resources/overview-diesel-exhaust-and-health [24]: #xref-ref-3-1 "View reference 3 in text" [25]: http://www.europarl.europa.eu/RegData/etudes/STUD/2016/587331/IPOL_STU(2016)587331_EN.pdf [26]: #xref-ref-4-1 "View reference 4 in text" [27]: http://www.dieselnet.com/standards/us/index.php [28]: #xref-ref-5-1 "View reference 5 in text" [29]: {openurl}?query=rft.jtitle%253DAnnu.%2BRev.%2BPublic%2BHealth%26rft.volume%253D41%26rft.spage%253D347%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [30]: #xref-ref-6-1 "View reference 6 in text" [31]: http://www.nytimes.com/2019/09/17/climate/trump-california-emissions-waiver.html [32]: #xref-ref-7-1 "View reference 7 in text" [33]: http://www.epa.gov/air-emissions-inventories/national-emissions-inventory-nei [34]: #xref-ref-8-1 "View reference 8 in text" [35]: {openurl}?query=rft.jtitle%253DAtmos.%2BEnviron.%26rft.volume%253D107%26rft.spage%253D70%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [36]: #xref-ref-9-1 "View reference 9 in text" [37]: http://www.healtheffects.org/system/files/GillilandRR190.pdf [38]: #xref-ref-10-1 "View reference 10 in text" [39]: #xref-ref-11-1 "View reference 11 in text" [40]: http://www.next10.org/publications/2017-gii [41]: #xref-ref-12-1 "View reference 12 in text" [42]: http://ww3.arb.ca.gov/board/books/2018/020818/18-1-2pres.pdf [43]: #xref-ref-13-1 "View reference 13 in text" [44]: {openurl}?query=rft.jtitle%253DScience%2BAdvances%26rft.stitle%253DSci%2BAdv%26rft.aulast%253DWu%26rft.auinit1%253DX.%26rft.volume%253D6%26rft.issue%253D45%26rft.spage%253Deabd4049%26rft.epage%253Deabd4049%26rft.atitle%253DAir%2Bpollution%2Band%2BCOVID-19%2Bmortality%2Bin%2Bthe%2BUnited%2BStates%253A%2BStrengths%2Band%2Blimitations%2Bof%2Ban%2Becological%2Bregression%2Banalysis%26rft_id%253Dinfo%253Adoi%252F10.1126%252Fsciadv.abd4049%26rft_id%253Dinfo%253Apmid%252F33148655%26rft.genre%253Darticle%26rft_val_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Ajournal%26ctx_ver%253DZ39.88-2004%26url_ver%253DZ39.88-2004%26url_ctx_fmt%253Dinfo%253Aofi%252Ffmt%253Akev%253Amtx%253Actx [45]: /lookup/ijlink/YTozOntzOjQ6InBhdGgiO3M6MTQ6Ii9sb29rdXAvaWpsaW5rIjtzOjU6InF1ZXJ5IjthOjQ6e3M6ODoibGlua1R5cGUiO3M6MzoiUERGIjtzOjExOiJqb3VybmFsQ29kZSI7czo4OiJhZHZhbmNlcyI7czo1OiJyZXNpZCI7czoxMzoiNi80NS9lYWJkNDA0OSI7czo0OiJhdG9tIjtzOjIzOiIvc2NpLzM3MS82NTM2LzEzMTQuYXRvbSI7fXM6ODoiZnJhZ21lbnQiO3M6MDoiIjt9 [46]: #xref-ref-14-1 "View reference 14 in text" [47]: #xref-ref-15-1 "View reference 15 in text" [48]: http://www.federalregister.gov/documents/2020/01/21/2020-00542/control-of-air-pollution-from-new-motor-vehicles-heavy-duty-engine-standards#citation-4-p3307

This post is a part of a series of posts that I will be making. You can read a more detailed version of this post on my personal blog by clicking here. Underneath you can see an overview of the series. We say that a model is linear if it's linear in the parameters not in the input variables. However, (1) is linear in both the parameters and the input variables, which limits it from adapting to nonlinear relationships.

We propose a novel estimator of the mutual information between two ordinal vectors $x$ and $y$. Our approach is inductive (as opposed to deductive) in that it depends on the data generating distribution solely through some nonparametric properties revealing associations in the data, and does not require having enough data to fully characterize the true joint distributions $P_{x, y}$. Specifically, our approach consists of (i) noting that $I\left(y; x\right) = I\left(u_y; u_x\right)$ where $u_y$ and $u_x$ are the copula-uniform dual representations of $y$ and $x$ (i.e. their images under the probability integral transform), and (ii) estimating the copula entropies $h\left(u_y\right)$, $h\left(u_x\right)$ and $h\left(u_y, u_x\right)$ by solving a maximum-entropy problem over the space of copula densities under a constraint of the type $\alpha_m = E\left[\phi_m(u_y, u_x)\right]$. We prove that, so long as the constraint is feasible, this problem admits a unique solution, it is in the exponential family, and it can be learned by solving a convex optimization problem. The resulting estimator, which we denote MIND, is marginal-invariant, always non-negative, unbounded for any sample size $n$, consistent, has MSE rate $O(1/n)$, and is more data-efficient than competing approaches. Beyond mutual information estimation, we illustrate that our approach may be used to mitigate mode collapse in GANs by maximizing the entropy of the copula of fake samples, a model we refer to as Copula Entropy Regularized GAN (CER-GAN).

With high forward gain, a negative feedback system has the ability to perform the inverse of a linear or non linear function that is in the feedback path. This property of negative feedback systems has been widely used in analog circuits to construct precise closed-loop functions. This paper describes how the property of a negative feedback system to perform inverse of a function can be used for training neural networks. This method does not require that the cost or activation functions be differentiable. Hence, it is able to learn a class of non-differentiable functions as well where a gradient descent-based method fails. We also show that gradient descent emerges as a special case of the proposed method. We have applied this method to the MNIST dataset and obtained results that shows the method is viable for neural network training. This method, to the best of our knowledge, is novel in machine learning.

Normalizing flow models have risen as a popular solution to the problem of density estimation, enabling high-quality synthetic data generation as well as exact probability density evaluation. However, in contexts where individuals are directly associated with the training data, releasing such a model raises privacy concerns. In this work, we propose the use of normalizing flow models that provide explicit differential privacy guarantees as a novel approach to the problem of privacy-preserving density estimation. We evaluate the efficacy of our approach empirically using benchmark datasets, and we demonstrate that our method substantially outperforms previous state-of-the-art approaches. We additionally show how our algorithm can be applied to the task of differentially private anomaly detection.

In a legal system, judgment consistency is regarded as one of the most important manifestations of fairness. However, due to the complexity of factual elements that impact sentencing in real-world scenarios, few works have been done on quantitatively measuring judgment consistency towards real-world data. In this paper, we propose an evaluation metric for judgment inconsistency, Legal Inconsistency Coefficient (LInCo), which aims to evaluate inconsistency between data groups divided by specific features (e.g., gender, region, race). We propose to simulate judges from different groups with legal judgment prediction (LJP) models and measure the judicial inconsistency with the disagreement of the judgment results given by LJP models trained on different groups. Experimental results on the synthetic data verify the effectiveness of LInCo. We further employ LInCo to explore the inconsistency in real cases and come to the following observations: (1) Both regional and gender inconsistency exist in the legal system, but gender inconsistency is much less than regional inconsistency; (2) The level of regional inconsistency varies little across different time periods; (3) In general, judicial inconsistency is negatively correlated with the severity of the criminal charges. Besides, we use LInCo to evaluate the performance of several de-bias methods, such as adversarial learning, and find that these mechanisms can effectively help LJP models to avoid suffering from data bias.

We study the estimation of causal parameters when not all confounders are observed and instead negative controls are available. Recent work has shown how these can enable identification and efficient estimation via two so-called bridge functions. In this paper, we tackle the primary challenge to causal inference using negative controls: the identification and estimation of these bridge functions. Previous work has relied on uniqueness and completeness assumptions on these functions that may be implausible in practice and also focused on their parametric estimation. Instead, we provide a new identification strategy that avoids both uniqueness and completeness. And, we provide a new estimators for these functions based on minimax learning formulations. These estimators accommodate general function classes such as reproducing Hilbert spaces and neural networks. We study finite-sample convergence results both for estimating bridge function themselves and for the final estimation of the causal parameter. We do this under a variety of combinations of assumptions that include realizability and closedness conditions on the hypothesis and critic classes employed in the minimax estimator. Depending on how much we are willing to assume, we obtain different convergence rates. In some cases, we show the estimate for the causal parameter may converge even when our bridge function estimators do not converge to any valid bridge function. And, in other cases, we show we can obtain semiparametric efficiency.

In this paper we apply the previously introduced approximation method based on the ANOVA (analysis of variance) decomposition and Grouped Transformations to synthetic and real data. The advantage of this method is the interpretability of the approximation, i.e., the ability to rank the importance of the attribute interactions or the variable couplings. Moreover, we are able to generate an attribute ranking to identify unimportant variables and reduce the dimensionality of the problem. We compare the method to other approaches on publicly available benchmark datasets.