United States

Loss landscape analysis is extremely useful for a deeper understanding of the generalization ability of deep neural network models. In this work, we propose a layerwise loss landscape analysis where the loss surface at every layer is studied independently and also on how each correlates to the overall loss surface. We study the layerwise loss landscape by studying the eigenspectra of the Hessian at each layer. In particular, our results show that the layerwise Hessian geometry is largely similar to the entire Hessian. We also report an interesting phenomenon where the Hessian eigenspectrum of middle layers of the deep neural network are observed to most similar to the overall Hessian eigenspectrum. We also show that the maximum eigenvalue and the trace of the Hessian (both full network and layerwise) reduce as training of the network progresses. We leverage on these observations to propose a new regularizer based on the trace of the layerwise Hessian. Penalizing the trace of the Hessian at every layer indirectly forces Stochastic Gradient Descent to converge to flatter minima, which are shown to have better generalization performance. In particular, we show that such a layerwise regularizer can be leveraged to penalize the middlemost layers alone, which yields promising results. Our empirical studies on well-known deep nets across datasets support the claims of this work.

AAAI 2021

A Deeper Look at the Hessian Eigenspectrum of Deep Neural Networks and Its Applications to Regularization

The purpose of the AAAI conference is to promote research in artificial intelligence (AI) and scientific exchange among AI researchers, practitioners, scientists, and engineers in affiliated disciplines. 

technical paper

Defect prediction, the task of predicting the presence of defects in source code artifacts, has broad application in software development. Defect prediction faces two major challenges, label scarcity, where only a small percentage of code artifacts are labeled, and data imbalance, where the majority of labeled artifacts are non-defective. Moreover, current defect prediction methods ignore the impact of information propagation among code artifacts and this negligence leads to performance degradation. In this paper, we propose DPCAG, a novel model to address the above three issues. We treat code artifacts as nodes in a graph, and learn to propagate influence among neighboring nodes iteratively in an EM framework. DPCAG dynamically adjusts the contributions of each node and selects high-confidence nodes for data augmentation. Experimental results on real-world benchmark datasets show that DPCAG improves performance compare to the state-of-the-art models. In particular, DPCAG achieves substantial performance superiority when measured by Matthews Correlation Coefficient (MCC), a metric that is widely acknowledged to be the most suitable for imbalanced data.

Towards Balanced Defect Prediction with Better information Propagation

Slot filling is a challenging task in Spoken Language Understanding (SLU). Supervised methods usually require large amounts of annotation to maintain desirable performance. A solution to relieve the heavy dependency on labeled data is to employ bootstrapping, which leverages unlabeled data. However, bootstrapping is known to suffer from semantic drift. We argue that semantic drift can be tackled by exploiting the correlation between slot values (phrases) and their respective types.  By using some particular weakly labeled data, namely the plain phrases included in sentences, we propose a weakly-supervised slot filling approach.  Our approach trains two models, namely a classifier and a tagger, which can effectively learn from each other on the weakly labeled data.  The experimental results demonstrate that our approach achieves better results than standard baselines on multiple datasets, especially in the low-resource setting.

Effective Slot Filling via Weakly-Supervised Dual-Model Learning

Automated data augmentation has shown superior performance in image recognition. Existing works search for dataset-level augmentation policies without considering individual sample variations, which are likely to be sub-optimal. On the other hand, learning different policies for different samples naively could greatly increase the computing cost. In this paper, we learn a sample-aware data augmentation policy efficiently by formulating it as a sample reweighting problem. Specifically, an augmentation policy network takes a transformation and the corresponding augmented image as inputs, and outputs a weight to adjust the augmented image loss computed by a task network. At training stage, the task network minimizes the weighted losses of augmented training images, while the policy network minimizes the loss of the task network on a validation set via meta-learning. We theoretically prove the convergence of the training procedure and further derive the exact convergence rate. Superior performance is achieved on widely-used benchmarks including CIFAR-10/100, Omniglot, and ImageNet.

MetaAugment: Sample-Aware Data Augmentation Policy Learning

Named entity recognition (NER) is a well-studied task in natural language processing. However, the widely-used sequence labeling framework is difficult to detect entities with nested structures. In this work, we view nested NER as constituency parsing with partially-observed trees and model it with partially-observed TreeCRFs. Specifically, we view all labeled entity spans as observed nodes in a constituency tree, and other spans as latent nodes. With the TreeCRF we achieve a uniform way to jointly model the observed and the latent nodes. To compute the probability of partial trees with partial marginalization, we propose a variant of the Inside algorithm, the Masked Inside algorithm, that supports different inference operations for different nodes (evaluation for the observed, marginalization for the latent, and rejection for nodes incompatible with the observed) with efficient parallelized implementation, thus significantly speeding up training and inference. Experiments show that our approach achieves the state-of-the-art (SOTA) F1 scores on the ACE2004, ACE2005 dataset, and shows comparable performance to SOTA models on the GENIA dataset. We release the code at https://github.com/FranxYao/Partially-Observed-TreeCRFs.

Nested Named Entity Recognition with Partially-Observed TreeCRFs

A rare disease is any disease that affects a very small percentage (1 in 1,500) of population. It is estimated that there are nearly 7,000 rare disease affecting 30 million patients in the U. S. alone. Most of the patients suffering from rare diseases experience multiple misdiagnoses and may never be diagnosed correctly. This is largely driven by the low prevalence of the disease that results in a lack of awareness among healthcare providers. There have been efforts from machine learning researchers to develop predictive models to help diagnose patients using healthcare datasets such as electronic health records and administrative claims. Most recently, transformer models have been applied to predict diseases BEHRT, G-BERT and Med-BERT. However, these have been developed specifically for electronic health records (EHR) and have not been designed to address rare disease challenges such as class imbalance, partial longitudinal data capture, and noisy labels. As a result, they deliver poor performance in predicting rare diseases compared with baselines. Besides, EHR datasets are generally confined to the hospital systems using them and do not capture a wider sample of patients thus limiting the availability of sufficient rare dis-ease patients in the dataset. To address these challenges, we introduced an extension of the BERT model tailored for rare disease diagnosis called RareBERT which has been trained on administrative claims datasets. RareBERT extends Med-BERT by including context embedding and temporal reference embedding. Moreover, we introduced a novel adaptive loss function to handle the class imbal-ance. In this paper, we show our experiments on diagnosing X-Linked Hypophosphatemia (XLH), a genetic rare disease. While RareBERT performs significantly better than the baseline models (79.9% AUPRC versus 30% AUPRC for Med-BERT), owing to the transformer architecture, it also shows its robustness in partial longitudinal data capture caused by poor capture of claims with a drop in performance of only 1.35% AUPRC, compared with 12% for Med-BERT and 33.0% for LSTM and 67.4% for boosting trees based baseline.

RareBERT: Transformer Architecture for Rare Disease Patient Identification Using Administrative Claims

Robin Hirsch posed in 1996 the Really Big Complexity Problem: classify the computational complexity of  the network satisfaction problem for all finite relation algebras A. We provide a complete classification for the case that A is symmetric and has a flexible atom; the problem is in this case NP-complete or in P. If a finite integral relation algebra has a flexible atom, then it has a normal representation B. We can then study the computational complexity of the network satisfaction problem of A using the universal-algebraic approach, via an analysis of the polymorphisms of B. We also use a Ramsey-type result of  NeÅ¡etÅ™il and RÃ¶dl and a complexity dichotomy result of Bulatov for conservative finite-domain constraint satisfaction problems.

Network Satisfaction for Symmetric Relation Algebras with a Flexible Atom

In this paper, we study the task of selecting the optimal response given a user and system utterance history in retrieval-based multi-turn dialog systems. Recently, pre-trained language models (e.g., BERT, RoBERTa, and ELECTRA) showed significant improvements in various natural language processing tasks. This and similar response selection tasks can also be solved using such language models by formulating the tasks as dialog--response binary classification tasks. Although existing works using this approach successfully obtained state-of-the-art results, we observe that language models trained in this manner tend to make predictions based on the relatedness of history and candidates, ignoring the sequential nature of multi-turn dialog systems. This suggests that the response selection task alone is insufficient for learning temporal dependencies between utterances. To this end, we propose utterance manipulation strategies (UMS) to address this problem. Specifically, UMS consist of several strategies (i.e., insertion, deletion, and search), which aid the response selection model towards maintaining dialog coherence. Further, UMS are self-supervised methods that do not require additional annotation and thus can be easily incorporated into existing approaches. Extensive evaluation across multiple languages and models shows that UMS are highly effective in teaching dialog consistency, which leads to models pushing the state-of-the-art with significant margins on multiple public benchmark datasets.

Do Response Selection Models Really Know What’s Next? Utterance Manipulation Strategies for Multi-Turn Response Selection

Open Set Domain Adaptation (OSDA) is a challenging domain adaptation setting which allows the existence of unknown classes on the target domain.
Although existing OSDA methods are good at classifying samples of known classes, they ignore the classification ability for the unknown samples, making them unbalanced OSDA methods.
To alleviate this problem, we propose a balanced OSDA methods which could recognize the unknown samples while maintain high classification performance for the known samples.
Specifically, to reduce the domain gaps, we first project the features to a hyperspherical latent space.
In this space, we propose to bound the centroid deviation angles to not only increase the intra-class compactness but also enlarge the inter-class margins.
With the bounded centroid deviation angles, we employ the statistical Extreme Value Theory to recognize the unknown samples that are misclassified into known classes.
In addition, to learn better centroids, we propose an improved centroid update strategy based on sample reweighting and adaptive update rate to cooperate with centroid alignment.
Experimental results on three OSDA benchmarks verify that our method can significantly outperform the compared methods and reduce the proportion of the unknown samples being misclassified into known classes.

Balanced Open Set Domain Adaptation via Centroid Alignment

We present an end-to-end joint training framework that explicitly models 6-DoF motion of multiple dynamic objects, ego-motion, and depth in a monocular camera setup without supervision. Our technical contributions are three-fold. First, we highlight the fundamental difference between inverse and forward projection while modeling the individual motion of each rigid object, and propose a geometrically correct projection pipeline using a neural forward projection module. Second, we design a unified instance-aware photometric and geometric consistency loss that holistically imposes self-supervisory signals for every background and object region. Lastly, we introduce a general-purpose auto-annotation scheme using any off-the-shelf instance segmentation and optical flow models to produce video instance segmentation maps that will be utilized as input to our training pipeline. These proposed elements are validated in a detailed ablation study. Through extensive experiments conducted on the KITTI and Cityscapes dataset, our framework is shown to outperform the state-of-the-art depth and motion estimation methods. Our code, dataset, and models are publicly available.

Learning Monocular Depth in Dynamic Scenes via Instance-Aware Projection Consistency

We consider the one-sided matching problem, where n agents have preferences over n items, and these preferences are induced by underlying cardinal valuation functions. The goal is to match every agent to a single item so as to maximize the social welfare. Most of the related literature, however, assumes that the values of the agents are not a priori known, and only access to the ordinal preferences of the agents over the items is provided. Consequently, this incomplete information leads to loss of efficiency, which is measured by the notion of distortion. In this paper, we further assume that the agents can answer a small number of queries, allowing us partial access to their values. We study the interplay between elicited cardinal information (measured by the number of queries per agent) and distortion for one-sided matching, as well as a wide range of well-studied related problems. Qualitatively, our results show that with a limited number of queries, it is possible to obtain significant improvements over the classic setting, where only access to ordinal information is given.

Downloads

Next from AAAI 2021

Towards Balanced Defect Prediction with Better information Propagation

Similar lecture

GO Hessian for Expectation-Based Objectives

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from AAAI 2021

Towards Balanced Defect Prediction with Better information Propagation

Similar lecture

GO Hessian for Expectation-Based Objectives

Downloads