An official website of the United States government
Deep neural networks (DNNs) have achieved near-human level accuracy on many datasets across different domains. But they are known to produce incorrect predictions with high confidence on inputs far from the training distribution. This challenge of lack of calibration of DNNs has limited the adoption of deep learning models in high-assurance systems such as autonomous driving, air traffic management, cybersecurity, and medical diagnosis. The problem of detecting when an input is outside the training distribution of a machine learning model, and hence, its prediction on this input cannot be trusted, has received significant attention recently. Several techniques based on statistical, geometric, topological, or relational signatures have been developed to detect the out-of-distribution (OOD) or novel inputs. In this paper, we present a runtime monitor based on predictive processing and dual process theory. We posit that the bottom-up deep neural networks can be monitored using top-down context models comprising two layers. The first layer is a feature density model that learns the joint distribution of the original DNN’s inputs, outputs, and the model’s explanation for its decisions. The second layer is a graph Markov neural network that captures an even broader context. We demonstrate the efficacy of our monitoring architecture in recognizing out-of-distribution and out-of-context inputs on the image classification and object detection tasks.
more »
« less
On Complementing End-To-End Human Behavior Predictors with Planning.
High capacity end-to-end approaches for human motion (behavior) prediction have the ability to represent subtle nuances in human behavior, but struggle with robustness to out of distribution inputs and tail events. Planning-based prediction, on the other hand, can reliably output decent-but-not-great predictions: it is much more stable in the face of distribution shift (as we verify in this work), but it has high inductive bias, missing important aspects that drive human decisions, and ignoring cognitive biases that make human behavior suboptimal. In this work, we analyze one family of approaches that strive to get the best of both worlds: use the end-to-end predictor on common cases, but do not rely on it for tail events / out-of-distribution inputs -- switch to the planning-based predictor there. We contribute an analysis of different approaches for detecting when to make this switch, using an autonomous driving domain. We find that promising approaches based on ensembling or generative modeling of the training distribution might not be reliable, but that there very simple methods which can perform surprisingly well -- including training a classifier to pick up on tell-tale issues in predicted trajectories.
more »
« less
- Award ID(s):
- 1652083
- PAR ID:
- 10315086
- Date Published:
- Journal Name:
- Robotics science and systems
- ISSN:
- 2330-765X
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Taking incompatible multiple drugs together may cause adverse interactions and side effects on the body. Accurate prediction of drug-drug interaction (DDI) events is essential for avoiding this issue. Recently, various artificial intelligence-based approaches have been proposed for predicting DDI events. However, DDI events are associated with complex relationships and mechanisms among drugs, targets, enzymes, transporters, molecular structures, etc. Existing approaches either partially or loosely consider these relationships and mechanisms by a non-end-to-end learning framework, resulting in sub-optimal feature extractions and fusions for prediction. Different from them, this paper proposes a Multimodal Knowledge Graph Fused End-to-end Neural Network (MKGFENN) that consists of two main parts: multimodal knowledge graph (MKG) and fused end-to-end neural network (FENN). First, MKG is constructed by comprehensively exploiting DDI events-associated relationships and mechanisms from four knowledge graphs of drugs-chemical entities, drug-substructures, drugs-drugs, and molecular structures. Correspondingly, a four channels graph neural network is designed to extract high-order and semantic features from MKG. Second, FENN designs a multi-layer perceptron to fuse the extracted features by end-to-end learning. With such designs, the feature extractions and fusions of DDI events are guaranteed to be comprehensive and optimal for prediction. Through extensive experiments on real drug datasets, we demonstrate that MKG-FENN exhibits high accuracy and significantly outperforms state-of-the-art models in predicting DDI events. The source code and supplementary file of this article are available on: https://github.com/wudi1989/MKG-FENN.more » « less
-
As technology advances, the need for safe, efficient, and collaborative human-robot-teams has become increasingly important. One of the most fundamental collaborative tasks in any setting is the object handover. Human-to-robot handovers can take either of two approaches: (1) direct hand-to-hand or (2) indirect hand-to-placement-to-pick-up. The latter approach ensures minimal contact between the human and robot but can also result in increased idle time due to having to wait for the object to first be placed down on a surface. To minimize such idle time, the robot must preemptively predict the human intent of where the object will be placed. Furthermore, for the robot to preemptively act in any sort of productive manner, predictions and motion planning must occur in real-time. We introduce a novel prediction-planning pipeline that allows the robot to preemptively move towards the human agent's intended placement location using gaze and gestures as model inputs. In this paper, we investigate the performance and drawbacks of our early intent predictor-planner as well as the practical benefits of using such a pipeline through a human-robot case study.more » « less
-
Skolnick, Jeffrey (Ed.)Systematically discovering protein-ligand interactions across the entire human and pathogen genomes is critical in chemical genomics, protein function prediction, drug discovery, and many other areas. However, more than 90% of gene families remain “dark”—i.e., their small-molecule ligands are undiscovered due to experimental limitations or human/historical biases. Existing computational approaches typically fail when the dark protein differs from those with known ligands. To address this challenge, we have developed a deep learning framework, called PortalCG, which consists of four novel components: (i) a 3-dimensional ligand binding site enhanced sequence pre-training strategy to encode the evolutionary links between ligand-binding sites across gene families; (ii) an end-to-end pretraining-fine-tuning strategy to reduce the impact of inaccuracy of predicted structures on function predictions by recognizing the sequence-structure-function paradigm; (iii) a new out-of-cluster meta-learning algorithm that extracts and accumulates information learned from predicting ligands of distinct gene families (meta-data) and applies the meta-data to a dark gene family; and (iv) a stress model selection step, using different gene families in the test data from those in the training and development data sets to facilitate model deployment in a real-world scenario. In extensive and rigorous benchmark experiments, PortalCG considerably outperformed state-of-the-art techniques of machine learning and protein-ligand docking when applied to dark gene families, and demonstrated its generalization power for target identifications and compound screenings under out-of-distribution (OOD) scenarios. Furthermore, in an external validation for the multi-target compound screening, the performance of PortalCG surpassed the rational design from medicinal chemists. Our results also suggest that a differentiable sequence-structure-function deep learning framework, where protein structural information serves as an intermediate layer, could be superior to conventional methodology where predicted protein structures were used for the compound screening. We applied PortalCG to two case studies to exemplify its potential in drug discovery: designing selective dual-antagonists of dopamine receptors for the treatment of opioid use disorder (OUD), and illuminating the understudied human genome for target diseases that do not yet have effective and safe therapeutics. Our results suggested that PortalCG is a viable solution to the OOD problem in exploring understudied regions of protein functional space.more » « less
-
Chen, J.Y.C. (Ed.)In recent years there has been a sharp increase in active shooter events, but there has been no introduction of new technology or tactics capable of increasing preparedness and training for active shooter events. This has raised a major concern about the lack of tools that would allow robust predictions of realistic human movements and the lack of understanding about the interaction in designated simulation environments. It is impractical to carry out live experiments where thousands of people are evacuated from buildings designed for every possible emergency condition. There has been progress in understanding human movement, human motion synthesis, crowd dynamics, indoor environments, and their relationships with active shooter events, but challenges remain. This paper presents a virtual reality (VR) experimental setup for conducting virtual evacuation drills in response to extreme events and demonstrates the behavior of agents during an active shooter environment. The behavior of agents is implemented using behavior trees in the Unity gaming engine. The VR experimental setup can simulate human behavior during an active shooter event in a campus setting. A presence questionnaire (PQ) was used in the user study to evaluate the effectiveness and engagement of our active shooter environment. The results show that majority of users agreed that the sense of presence was increased when using the emergency response training environment for a building evacuation environment.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.15607/RSS.2021.XVII.037
