Browsing by Author "Yang, Fan"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item Counterfactuals for Interpretable Machine Learning: Model Reasoning from “What” to “How”(2023-05-23) Yang, Fan; Hu, XiaWith the extensive usage of machine learning (ML) in real-world applications, how to effectively explain the behaviors of ML models is becoming increasingly significant. A bunch of interpretation techniques have then been proposed, aiming to facilitate end-users for a better understanding towards the model working mechanism. Existing techniques for interpretable machine learning mainly focus on the feature attribution methods, where highly contributed features are exported as evidence for model predictions. However, those obtained feature contribution scores are not discriminative in nature, which makes them limited in reasoning decisions and understanding "how". Counterfactual Explanation, serving as one of the emerging types of ML interpretations, has raised the attention from both researchers and practitioners in recent years. Counterfactual explanation is essentially a series of hypothetical data samples, which is categorized under the example-based reasoning methodology and explored under "what-if" circumstances. The overall interpretation goal of counterfactuals is to indicate how the model decision alters with input perturbations. With valid counterfactual explanations, end-users can know how to flip the model decisions to a preferred outcome, so as to get a better sense of the decision boundaries. In this thesis, I will cover my previous research efforts on counterfactual explanations, and outline the introduction from three different perspectives. Firstly, for counterfactual derivation, I designed a framework to generate counterfactuals specifically for raw data instances with the proposed Attribute-Informed Perturbation. By utilizing generative models conditioned with different attributes, counterfactuals with desired labels can be obtained effectively. Instead of directly modifying instances in the data space, I iteratively optimized the constructed attribute-informed latent space, where features are more robust and semantic. Secondly, for counterfactual explainer deployment, I proposed a Model-based Counterfactual Synthesizer framework for efficient interpretation. I analyzed the model-based counterfactual process, and constructed a base synthesizer by adopting the conditional generative adversarial net structure. To better approximate the counterfactual universe for those minor queries, I employed the umbrella sampling technique to conduct the synthesizer training. I also enhanced the synthesizer by incorporating the causal dependence among attributes, and further validated its correctness through the causality identification approach. Thirdly, for counterfactual delivery to stake-holders, I proposed a novel framework to generate differentially private counterfactuals, where noises are injected for protection while maintaining the explanation roles. I trained an autoencoder with the functional mechanism to construct noisy class prototypes, and then derived the counterfactual explanation from the latent prototypes based on the post-processing immunity of differential privacy. Beyond general stake-holders, I also specifically proposed two explanation delivery frameworks for end-users and model developers. The further research goals are to focus on the sequential counterfactual which is more actionable for end-users, and the global counterfactual which is more insightful for model developers. At the end of thesis, I will list several promising directions to explore in the future.Item Germanium/silicon of the Ediacaran-Cambrian Laobao cherts: Implications for the bedded chert formation and paleoenvironment interpretations(American Geophysical Union, 2015) Dong, Lin; Shen, Bing; Lee, Cin-Ty A.; Shu, Xu-jie; Peng, Yang; Sun, Yuanlin; Tang, Zhuanhong; Rong, Hong; Lang, Xianguo; Ma, Haoran; Yang, Fan; Guo, WenSedimentary strata of the terminal Ediacaran (635–542 Ma) to early Cambrian (542–488 Ma) Laobao-Liuchapo bedded cherts in the South China Block include the Ediacaran Oxidation Event and the Cambrian explosion. Understanding the origin and depositional environment of the bedded cherts may provide insight into how the Earth's surface environment changed between the Proterozoic and Phanerozoic. We measured major and trace element compositions and Ge/Si ratios of the Laobao cherts from northern Guangxi province. The Laobao cherts were deposited in the deep basinal environment of the South China Block. We show that the composition of the Laobao cherts is determined by a mixture of four components: quartz, clay, carbonate, and pyrite/iron-oxide. The quartz component is the dominant component of the Laobao cherts. The maximum estimated Ge/Si of the quartz component is between 0.4 and 0.5 μmol/mol, which is close to the Ge/Si of modern seawater and biogenic silica but 1 order of magnitude lower than that of hydrothermal fluids. These Ge/Si systematics suggest that normal seawater rather than mid-ocean ridge hydrothermal fluids is the primary Si source for the Laobao cherts. The Ge/Si of the clay component varies between 1 and 10 μmol/mol, which is comparable to the Ge/Si of typical marine clays, but 10–100 times lower than that of chert nodules from early Ediacaran beds (the Doushantuo Formation) predating the terminal Ediacaran Laobao cherts studied here. Our observations indicate that the clay component Ge/Si ratio decreased from the early Ediacaran to the late Ediacaran. We speculate that high Ge/Si ratios in clays reflect the preferential chelation of Ge by dissolved organic compounds adsorbed onto clays. If so, this suggests that the decrease in Ge/Si ratio of the clay component in the Ediacaran signifies a decrease in the total dissolved organic carbon content of seawater toward the Ediacaran-Cambrian transition, consistent with oxidation of the oceans during the late Ediacaran. Finally, the seawater origin of the Laobao cherts also suggests that replacement of carbonate may not be the primary cause for bedded chert formation. Instead, direct precipitation from seawater or early diagenetic silicification of calcareous sediments, perhaps due to the emergence of Si-accumulation bacteria, may have been responsible for the bedded Laobao-Liuchapo chert formation in South China Block.Item The molecular structure of green fluorescent protein(1997) Yang, Fan; Phillips, George N., Jr.The crystal structure of recombinant wild-type green fluorescent protein (GFP) has been solved to a resolution of 1.9 A by multiwavelength anomalous dispersion phasing methods using selenomethionyl GFP crystals. The protein is in the shape of a cylinder, comprising 11 strands of $\beta$-sheet with an $\alpha$-helix inside and short helical segments on the ends of the cylinder. This motif, with $\beta$-structure on the outside and $\alpha$-helix on the inside, represents a new protein fold. Two protomers pack closely together to form a dimer in the crystal. The fluorophores are protected inside the cylinders, and their structures are consistent with the formation of aromatic systems made up of Tyr66 with oxidation of its C$\sb{\alpha}$-C$\sb{\beta}$ bond coupled with cyclization of the neighboring glycine and serine residues. The environment inside the cylinder explains the effects of many existing mutants of GFP and suggests which side chains could be modified to change the spectral properties of GFP. Furthermore, the identification of the dimer contacts may allow mutagenic control of the state of assembly of the protein. GFP can be reduced by sodium dithionite and as a result, loses its fluorescence. The structure of reduced GFP has been solved which shows that the side chain of Tyr66 at the fluorophore is the group being reduced. While oxidized, the C$\sb{\alpha}$ atom of Tyr66 is not a chiral center since all fluorophore atoms are co-planar. After reduction, the C$\sb{\alpha}$ atom returns to a chiral center, but has either an L or D configuration, indicating that the breakage of the resonance system eliminates the fluorescence of GFP and the reduction of Tyr66 is not stereospecific.