Missing data is a pervasive and challenging issue in various applications of statistical inference, such as healthcare, economics, and the social sciences. Data are said to be Missing at Random (MAR) when the mechanism of missingness depends only on the observed data. Strategies to deal with MAR have been extensively investigated in the literature (Dempster et al., 1977; Robins et al., 1994; Tsiatis, 2006; Little and Rubin, 2019). In many practical settings, MAR is not a realistic assumption. Instead, missingness often depends on variables that are themselves missing. Such settings are said to exhibit nonignorable missingness, with the resulting data being Missing Not at Random (MNAR) (Fielding et al., 2008; Schafer and Graham, 2002), A classic example of a scenario with MNAR data occurs in longitudinal studies, due to the treatment's toxicity, some patients may become too ill to visit the clinic, leading to the situation where the outcome of certain patients with circumstances associated with those outcomes are more likely to be lost to follow-up (Ibrahim et al., 2012). Previous MNAR models have typically imposed constraints on the target distribution and its missingness mechanism, ensuring the parameter of interest can be identified. This approach goes back to the work of Heckman (1979), who proposed an outcome-selection model based on parametric modeling of the outcome variable and missing pattern. Little (1993) introduced the pattern-mixture model where one needs to specify the distribution for each missing data pattern independently.

Learning with noisy label (LNL) is a classic problem that has been extensively studied for image tasks, but much less for video in the literature. A straightforward migration from images to videos without considering the properties of videos, such as computational cost and redundant information, is not a sound choice. In this paper, we propose two new strategies for video analysis with noisy labels: 1) A lightweight channel selection method dubbed as Channel Truncation for feature-based label noise detection. This method selects the most discriminative channels to split clean and noisy instances in each category; 2) A novel contrastive strategy dubbed as Noise Contrastive Learning, which constructs the relationship between clean and noisy instances to regularize model training. Experiments on three well-known benchmark datasets for video classification show that our proposed tru{\bf N}cat{\bf E}-split-contr{\bf A}s{\bf T} (NEAT) significantly outperforms the existing baselines. By reducing the dimension to 10\% of it, our method achieves over 0.4 noise detection F1-score and 5\% classification accuracy improvement on Mini-Kinetics dataset under severe noise (symmetric-80\%). Thanks to Noise Contrastive Learning, the average classification accuracy improvement on Mini-Kinetics and Sth-Sth-V1 is over 1.6\%.

In this paper, we propose a Graph Inception Diffusion Networks(GIDN) model. This model generalizes graph diffusion in different feature spaces, and uses the inception module to avoid the large amount of computations caused by complex network structures. We evaluate GIDN model on Open Graph Benchmark(OGB) datasets, reached an 11% higher performance than AGDN on ogbl-collab dataset.