Search for: All records

Federated learning (FL) is a learning paradigm that allows the central server to learn from different data sources while keeping the data private locally. Without controlling and monitoring the local data collection process, the locally available training labels are likely noisy, i.e., the collected training labels differ from the unobservable ground truth. Additionally, in heterogenous FL, each local client may only have access to a subset of label space (referred to as openset label learning), meanwhile without overlapping with others. In this work, we study the challenge of FL with local openset noisy labels. We observe that many existing solutions in the noisy label literature, e.g., loss correction, are ineffective during local training due to overfitting to noisy labels and being not generalizable to openset labels. For the methods in FL, different estimated metrics are shared. To address the problems, we design a label communication mechanism that shares "contrastive labels" randomly selected from clients with the server. The privacy of the shared contrastive labels is protected by label differential privacy (DP). Both the DP guarantee and the effectiveness of our approach are theoretically guaranteed. Compared with several baseline methods, our solution shows its efficiency in several public benchmarks and real-world datasets under different noise ratios and noise models. 

ABSTRACT Identifying species with disproportionate effects on other species under press perturbations is essential, yet how species traits and community context drive their ‘keystone‐ness’ remain unclear. We quantified keystone‐ness as linearly approximated per capita net effect derived from normalised inverse community matrices and as non‐linear per capita community biomass change from simulated perturbations in food webs with varying biomass structure. In bottom‐heavy webs (negative relationship between species' body mass and their biomass within the web), larger species at higher trophic levels tended to be keystone species, whereas in top‐heavy webs (positive body mass to biomass relationship), the opposite was true and the relationships between species' energetic traits and keystone‐ness were weakened or reversed compared to bottom‐heavy webs. Linear approximations aligned well with non‐linear responses in bottom‐heavy webs, but were less consistent in top‐heavy webs. These findings highlight the importance of community context in shaping species' keystone‐ness and informing effective conservation actions. 

Gas-phase oxygenated organic molecules (OOMs) can contribute significantly to both atmospheric new particle growth and secondary organic aerosol formation. Precursor apportionment of atmospheric OOMs connects them with volatile organic compounds (VOCs). Since atmospheric OOMs are often highly functionalized products of multistep reactions, it is challenging to reveal the complete mapping relationships between OOMs and their precursors. In this study, we demonstrate that the machine learning method is useful in attributing atmospheric OOMs to their precursors using several chemical indicators, such as O/C ratio and H/C ratio. The model is trained and tested using data acquired in controlled laboratory experiments, covering the oxidation products of four main types of VOCs (isoprene, monoterpenes, aliphatics, and aromatics). Then, the model is used for analyzing atmospheric OOMs measured in both urban Beijing and a boreal forest environment in southern Finland. The results suggest that atmospheric OOMs in these two environments can be reasonably assigned to their precursors. Beijing is an anthropogenic VOC dominated environment with ∼64% aromatic and aliphatic OOMs, and the other boreal forested area has ∼76% monoterpene OOMs. This pilot study shows that machine learning can be a promising tool in atmospheric chemistry for connecting the dots. 

Federated Learning (FL) is an emerging learning scheme that allows different distributed clients to train deep neural networks together without data sharing. Neural networks have become popular due to their unprecedented success. To the best of our knowledge, the theoretical guarantees of FL concerning neural networks with explicit forms and multi-step updates are unexplored. Nevertheless, training analysis of neural networks in FL is non-trivial for two reasons: first, the objective loss function we are optimizing is non-smooth and non-convex, and second, we are even not updating in the gradient direction. Existing convergence results for gradient descent-based methods heavily rely on the fact that the gradient direction is used for updating. The current paper presents a new class of convergence analysis for FL, Federated Neural Tangent Kernel (FL-NTK), which corresponds to overparamterized ReLU neural networks trained by gradient descent in FL and is inspired by the analysis in Neural Tangent Kernel (NTK). Theoretically, FL-NTK converges to a global-optimal solution at a linear rate with properly tuned learning parameters. Furthermore, with proper distributional assumptions, FL-NTK can also achieve good generalization. The proposed theoretical analysis scheme can be generalized to more complex neural networks. 

Federated Learning (FL) is an emerging learning scheme that allows different distributed clients to train deep neural networks together without data sharing. Neural networks have become popular due to their unprecedented success. To the best of our knowledge, the theoretical guarantees of FL concerning neural networks with explicit forms and multi-step updates are unexplored. Nevertheless, training analysis of neural networks in FL is non-trivial for two reasons: first, the objective loss function we are optimizing is non-smooth and non-convex, and second, we are even not updating in the gradient direction. Existing convergence results for gradient descent-based methods heavily rely on the fact that the gradient direction is used for updating. This paper presents a new class of convergence analysis for FL, Federated Learning Neural Tangent Kernel (FL-NTK), which corresponds to over-paramterized ReLU neural networks trained by gradient descent in FL and is inspired by the analysis in Neural Tangent Kernel (NTK). Theoretically, FL-NTK converges to a global-optimal solution at a linear rate with properly tuned learning parameters. Furthermore, with proper distributional assumptions, FL-NTK can also achieve good generalization. 

Abstract. It has been widely observed around the world that the frequency and intensityof new particle formation (NPF) events are reduced during periods of highrelative humidity (RH). The current study focuses on how RH affects theformation of highly oxidized molecules (HOMs), which are key components ofNPF and initial growth caused by oxidized organics. The ozonolysis ofα-pinene, limonene, and Δ³-carene, with and without OHscavengers, were carried out under low NO_x conditions undera range of RH (from ∼3 % to ∼92 %) in atemperature-controlled flow tube to generate secondary organic aerosol (SOA).A Scanning Mobility Particle Sizer (SMPS) was used to measure the sizedistribution of generated particles, and a novel transverse ionizationchemical ionization inlet with a high-resolution time-of-fight massspectrometer detected HOMs. A major finding from this work is that neitherthe detected HOMs nor their abundance changed significantly with RH, whichindicates that the detected HOMs must be formed from water-independentpathways. In fact, the distinguished OH- and O₃-derived peroxyradicals (RO₂), HOM monomers, and HOM dimers could mostly beexplained by the autoxidation of RO₂ followed by bimolecularreactions with other RO₂ or hydroperoxy radicals (HO₂),rather than from a water-influenced pathway like through the formation of astabilized Criegee intermediate (sCI). However, as RH increased from ∼3 % to ∼92 %, the total SOA number concentrations decreased bya factor of 2–3 while SOA mass concentrations remained relatively constant. These observations show that, whilehigh RH appears to inhibit NPF as evident by the decreasing numberconcentration, this reduction is not caused by a decrease inRO₂-derived HOM formation. Possible explanations for these phenomenawere discussed.