Hierarchical Knowledge Propagation and Distillation for Few-Shot Learning

Recent research efforts on Few-Shot Learning (FSL) have achieved extensive progress. However, the existing efforts primarily focus on the transductive setting of FSL, which is heavily challenged by the limited quantity of the unlabeled query set. Although a few inductive-based FSL methods have been studied, most of them emphasize learning superb feature extraction networks. As a result, they may ignore the relations between sample-level and class-level representations, which are particularly crucial when labeled samples are scarce. This paper proposes an inductive FSL framework that leverages the Hierarchical Knowledge Propagation and Distillation, named HKPD. To learn more discriminative sample-level representations, HKPD first constructs a sample-level information propagation module that explores pairwise sample relations. Subsequently, a class-level information propagation module is designed to obtain and update the class-level information. Moreover, a self-distillation module is adopted to further improve the learned representations by propagating the obtained knowledge across this hierarchical architecture. Extensive experiments conducted on the commonly used few-shot benchmark datasets demonstrate the superiority of the proposed HKPD method, which outperforms the current state-of-the-art methods.

Interleaving Effects in Blindfolded Perceptual Learning Across Various Sensory Modalities

Research on sequence effects on learning visual categories has shown that interleaving (i.e., studying the categories in a mixed manner) facilitates category induction as compared to blocking (i.e., studying the categories one by one), but learners are unaware of the interleaving effect and prefer blocking. However, little attention has been paid to sequence effects in perceptual learning across further sensory modalities. The present (preregistered) research addresses this shortcoming by using auditory (birdcalls), olfactory (tealeaves), gustatory (ingredient mixtures), and tactile (stones) stimuli across four analog experiments. The number of participants per experiment was determined based on a medium effect size of interleaving. Participants studied six categories (with six exemplars, respectively) either interleaved or blocked. No single experiment showed a significant effect of interleaving. We ran a comprehensive meta-analysis based on the data from all experiments, which revealed a significant small effect of interleaving, demonstrating its applicability to perceptual learning across all sensory modalities. Learners in the interleaved condition underestimated their classification performance. Overall, learners did not rate interleaving as less effective than blocking, which is at odds with previous studies that consistently demonstrated a metacognitive preference of blocking. Our findings suggest that learners rely less on conventional beliefs about the effectivity of study sequence when dealing with unfamiliar (blindfolded) perceptual learning tasks.

Inductive inference of gene regulatory network using supervised and semi-supervised graph neural networks

•

A novel graph classification formulation in inferring gene regulatory relationships.

•

Graph neural network is powerful to ensemble powers from heuristic skeletons.

•

Our results show GRGRNN outperforms previous methods inductively on benchmarks.

•

GRGNN can be interpreted following the biological network motif hypothesis.

Discovering gene regulatory relationships and reconstructing gene regulatory networks (GRN) based on gene expression data is a classical, long-standing computational challenge in bioinformatics. Computationally inferring a possible regulatory relationship between two genes can be formulated as a link prediction problem between two nodes in a graph. Graph neural network (GNN) provides an opportunity to construct GRN by integrating topological neighbor propagation through the whole gene network. We propose an end-to-end gene regulatory graph neural network (GRGNN) approach to reconstruct GRNs from scratch utilizing the gene expression data, in both a supervised and a semi-supervised framework. To get better inductive generalization capability, GRN inference is formulated as a graph classification problem, to distinguish whether a subgraph centered at two nodes contains the link between the two nodes. A linked pair between a transcription factor (TF) and a target gene, and their neighbors are labeled as a positive subgraph, while an unlinked TF and target gene pair and their neighbors are labeled as a negative subgraph. A GNN model is constructed with node features from both explicit gene expression and graph embedding. We demonstrate a noisy starting graph structure built from partial information, such as Pearson’s correlation coefficient and mutual information can help guide the GRN inference through an appropriate ensemble technique. Furthermore, a semi-supervised scheme is implemented to increase the quality of the classifier. When compared with established methods, GRGNN achieved state-of-the-art performance on the DREAM5 GRN inference benchmarks. GRGNN is publicly available at https://github.com/juexinwang/GRGNN.

Stable solution to l 2,1-based robust inductive matrix completion and its application in linking long noncoding RNAs to human diseases

Backgrounds

A large number of long intergenic non-coding RNAs (lincRNAs) are linked to a broad spectrum of human diseases. The disease association with many other lincRNAs still remain as puzzle. Validation of such links between the two entities through biological experiments are expensive. However, a plethora lincRNA-data are available now, thanks to the High Throughput Sequencing (HTS) platforms, Genome Wide Association Studies (GWAS), etc, which opens the opportunity for cutting-edge machine learning and data mining approaches to extract meaningful relationships among lincRNAs and diseases. However, there are only a few in silico lincRNA-disease association inference tools available to date, and none of them utilizes side information of both the entities simultaneously in a single framework.

Methods

The recently developed Inductive Matrix Completion (IMC) technique provides a recommendation platform among two entities considering respective side information about them. However, the formulation of IMC is incapable of handling noise and outliers that may be present in the datasets, while data sparsity consideration is another issue with the standard IMC method. Thus, a robust version of IMC is needed that can solve the two issues. As a remedy, in this paper, we propose Stable Robust Inductive Matrix Completion (SRIMC) that utilizes the l_2,1 norm based regularization to optimize the objective function with a unique 2-step stable solution approach.

Results

We applied SRIMC to the available association data between human lincRNAs and OMIM disease phenotypes as well as a diverse set of side information about the lincRNAs and the diseases. The method performs better than the state-of-the-art methods in terms of precision@k and recall@k at the top-k disease prioritization to the subject lincRNAs. We also demonstrate that SRIMC is equally effective for querying about novel lincRNAs, as well as predicting rank of a newly known disease for a set of well-characterized lincRNAs.

Conclusions

With the experimental results and computational evaluation, we show that SRIMC is robust in handling datasets with noise and outliers as well as dealing with novel lincRNAs and disease phenotypes.