One of the central elements of any causal inference is an object called structural causal model (SCM), which represents a collection of mechanisms and exogenous sources of random variation of the system under investigation (Pearl, 2000). An important property of many kinds of neural networks is universal approximability: the ability to approximate any function to arbitrary precision. Given this property, one may be tempted to surmise that a collection of neural nets is capable of learning any SCM by training on data generated by that SCM. In this paper, we show this is not the case by disentangling the notions of expressivity and learnability. Specifically, we show that the causal hierarchy theorem (Thm. 1, Bareinboim et al., 2020), which describes the limits of what can be learned from data, still holds for neural models. For instance, an arbitrarily complex and expressive neural net is unable to predict the effects of interventions given observational data alone. Given this result, we introduce a special type of SCM called a neural causal model (NCM), and formalize a new type of inductive bias to encode structural constraints necessary for performing causal inferences. Building on this new class of models, we focus on solving two canonical tasks found in the literature known as causal identification and estimation. Leveraging the neural toolbox, we develop an algorithm that is both sufficient and necessary to determine whether a causal effect can be learned from data (i.e., causal identifiability); it then estimates the effect whenever identifiability holds (causal estimation). Simulations corroborate the proposed approach.
more »
« less
Topological Detection of Trojaned Neural Networks
Deep neural networks are known to have security issues. One particular threat is the Trojan attack. It occurs when the attackers stealthily manipulate the model's behavior through Trojaned training samples, which can later be exploited.
Guided by basic neuroscientific principles we discover subtle -- yet critical -- structural deviation characterizing Trojaned models. In our analysis we use topological tools. They allow us to model high-order dependencies in the networks, robustly compare different networks, and localize structural abnormalities. One interesting observation is that Trojaned models develop short-cuts from input to output layers.
Inspired by these observations, we devise a strategy for robust detection of Trojaned models. Compared to standard baselines it displays better performance on multiple benchmarks.
more »
« less
- Award ID(s):
- 1910873
- NSF-PAR ID:
- 10366262
- Date Published:
- Journal Name:
- Advances in neural information processing systems
- ISSN:
- 1049-5258
- Page Range / eLocation ID:
- 17258--17272
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
more » « less
Abstract Background The recent development of high-throughput sequencing has created a large collection of multi-omics data, which enables researchers to better investigate cancer molecular profiles and cancer taxonomy based on molecular subtypes. Integrating multi-omics data has been proven to be effective for building more precise classification models. Most current multi-omics integrative models use either an early fusion in the form of concatenation or late fusion with a separate feature extractor for each omic, which are mainly based on deep neural networks. Due to the nature of biological systems, graphs are a better structural representation of bio-medical data. Although few graph neural network (GNN) based multi-omics integrative methods have been proposed, they suffer from three common disadvantages. One is most of them use only one type of connection, either inter-omics or intra-omic connection; second, they only consider one kind of GNN layer, either graph convolution network (GCN) or graph attention network (GAT); and third, most of these methods have not been tested on a more complex classification task, such as cancer molecular subtypes.
Results In this study, we propose a novel end-to-end multi-omics GNN framework for accurate and robust cancer subtype classification. The proposed model utilizes multi-omics data in the form of heterogeneous multi-layer graphs, which combine both inter-omics and intra-omic connections from established biological knowledge. The proposed model incorporates learned graph features and global genome features for accurate classification. We tested the proposed model on the Cancer Genome Atlas (TCGA) Pan-cancer dataset and TCGA breast invasive carcinoma (BRCA) dataset for molecular subtype and cancer subtype classification, respectively. The proposed model shows superior performance compared to four current state-of-the-art baseline models in terms of accuracy, F1 score, precision, and recall. The comparative analysis of GAT-based models and GCN-based models reveals that GAT-based models are preferred for smaller graphs with less information and GCN-based models are preferred for larger graphs with extra information.
-
Graph neural networks (GNNs) have achieved tremendous success on multiple graph-based learning tasks by fusing network structure and node features. Modern GNN models are built upon iterative aggregation of neighbor's/proximity features by message passing. Its prediction performance has been shown to be strongly bounded by assortative mixing in the graph, a key property wherein nodes with similar attributes mix/connect with each other. We observe that real world networks exhibit heterogeneous or diverse mixing patterns and the conventional global measurement of assortativity, such as global assortativity coefficient, may not be a representative statistic in quantifying this mixing. We adopt a generalized concept, node-level assortativity, one that is based at the node level to better represent the diverse patterns and accurately quantify the learnability of GNNs. We find that the prediction performance of a wide range of GNN models is highly correlated with the node level assortativity. To break this limit, in this work, we focus on transforming the input graph into a computation graph which contains both proximity and structural information as distinct type of edges. The resulted multi-relational graph has an enhanced level of assortativity and, more importantly, preserves rich information from the original graph. We then propose to run GNNs on this computation graph and show that adaptively choosing between structure and proximity leads to improved performance under diverse mixing. Empirically, we show the benefits of adopting our transformation framework for semi-supervised node classification task on a variety of real world graph learning benchmarks.more » « less
-
more » « less
Abstract Predicting residue‐residue distance relationships (eg, contacts) has become the key direction to advance protein structure prediction since 2014 CASP11 experiment, while deep learning has revolutionized the technology for contact and distance distribution prediction since its debut in 2012 CASP10 experiment. During 2018 CASP13 experiment, we enhanced our MULTICOM protein structure prediction system with three major components: contact distance prediction based on deep convolutional neural networks, distance‐driven template‐free (ab initio) modeling, and protein model ranking empowered by deep learning and contact prediction. Our experiment demonstrates that contact distance prediction and deep learning methods are the key reasons that MULTICOM was ranked 3rd out of all 98 predictors in both template‐free and template‐based structure modeling in CASP13. Deep convolutional neural network can utilize global information in pairwise residue‐residue features such as coevolution scores to substantially improve contact distance prediction, which played a decisive role in correctly folding some free modeling and hard template‐based modeling targets. Deep learning also successfully integrated one‐dimensional structural features, two‐dimensional contact information, and three‐dimensional structural quality scores to improve protein model quality assessment, where the contact prediction was demonstrated to consistently enhance ranking of protein models for the first time. The success of MULTICOM system clearly shows that protein contact distance prediction and model selection driven by deep learning holds the key of solving protein structure prediction problem. However, there are still challenges in accurately predicting protein contact distance when there are few homologous sequences, folding proteins from noisy contact distances, and ranking models of hard targets.
-
more » « less
Abstract Data-driven generative design (DDGD) methods utilize deep neural networks to create novel designs based on existing data. The structure-aware DDGD method can handle complex geometries and automate the assembly of separate components into systems, showing promise in facilitating creative designs. However, determining the appropriate vectorized design representation (VDR) to evaluate 3D shapes generated from the structure-aware DDGD model remains largely unexplored. To that end, we conducted a comparative analysis of surrogate models’ performance in predicting the engineering performance of 3D shapes using VDRs from two sources: the trained latent space of structure-aware DDGD models encoding structural and geometric information and an embedding method encoding only geometric information. We conducted two case studies: one involving 3D car models focusing on drag coefficients and the other involving 3D aircraft models considering both drag and lift coefficients. Our results demonstrate that using latent vectors as VDRs can significantly deteriorate surrogate models’ predictions. Moreover, increasing the dimensionality of the VDRs in the embedding method may not necessarily improve the prediction, especially when the VDRs contain more information irrelevant to the engineering performance. Therefore, when selecting VDRs for surrogate modeling, the latent vectors obtained from training structure-aware DDGD models must be used with caution, although they are more accessible once training is complete. The underlying physics associated with the engineering performance should be paid attention. This paper provides empirical evidence for the effectiveness of different types of VDRs of structure-aware DDGD for surrogate modeling, thus facilitating the construction of better surrogate models for AI-generated designs.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
