An official website of the United States government
In today's world, AI systems need to make sense of large amounts of data as it unfolds in real-time, whether it's a video from surveillance and monitoring cameras, streams of egocentric footage, or sequences in other domains such as text or audio. The ability to break these continuous data streams into meaningful events, discover nested structures, and predict what might happen next at different levels of abstraction is crucial for applications ranging from passive surveillance systems to sensory-motor autonomous learning. However, most existing models rely heavily on large, annotated datasets with fixed data distributions and offline epoch-based training, which makes them impractical for handling the unpredictability and scale of dynamic real-world environments. This dissertation tackles these challenges by introducing a set of predictive models designed to process streaming data efficiently, segment events, and build sequential memory models without supervision or data storage. First, we present a single-layer predictive model that segments long, unstructured video streams by detecting temporal events and spatially localizing objects in each frame. The model is applied to wildlife monitoring footage, where it processes continuous, high-frame-rate video and successfully detects and tracks events without supervision. It operates in an online streaming manner to perform simultaneous training and inference without storing or revisiting the processed data. This approach alleviates the need for manual labeling, making it ideal for handling long-duration, real-world video footage. Building on this, we introduce STREAMER, a multi-layered architecture that extends the single-layer model into a hierarchical predictive framework. STREAMER segments events at different levels of abstraction, capturing the compositional structure of activities in egocentric videos. By dynamically adapting to various timescales, it creates a hierarchy of nested events and forms more complex and abstract representations of the input data. Finally, we propose the Predictive Attractor Model (PAM), which builds biologically plausible memory models of sequential data. Inspired by neuroscience, PAM uses sparse distributed representations and local learning rules to avoid catastrophic forgetting, allowing it to continually learn and make predictions without overwriting previous knowledge. Unlike many traditional models, PAM can generate multiple potential future outcomes conditioned on the same context, which allows for handling uncertainty in generative tasks. Together, these models form a unified framework of predictive learning that addresses multiple challenges in event understanding and temporal data analyses. By using prediction as the core mechanism, they segment continuous data streams into events, discover hierarchical structures across multiple levels of abstraction, learn semantic event representations, and model sequences without catastrophic forgetting.
more »
« less
Hierarchical Tree-Based Sequential Event Prediction with Application in the Aviation Accident Report
Sequential event prediction is a well-studied area and has been widely used in proactive management, recommender systems and healthcare. One major assumption of the existing sequential event prediction methods is that similar event sequence patterns in the historical record will repeat themselves, enabling us to predict future events. However, in reality, the assumption becomes less convincing when we are trying to predict rare or unique sequences. Furthermore, the representation of the event may be complex with hierarchical structures. In this paper, we aim to solve this issue by taking advantage of the multi-level or hierarchical representation of these rare events. We proposed to build a sequential Encoder-Decoder framework to predict the event sequences. More specifically, in the encoding layer, we built a hierarchical embedding representation for the events. In the decoding layer, we first predict the high-level events and the low-level events are generated according to a hierarchical graphical structure. We propose to link the encoding decoding layers with the temporal models for future event prediction. In this article, we further discussed applying the proposed model into the failure event prediction according to the aviation accident reports and have shown improved accuracy and model interpretability.
more »
« less
- Award ID(s):
- 1830363
- PAR ID:
- 10291555
- Date Published:
- Journal Name:
- Proceedings - 2021 IEEE 37th International Conference on Data Engineering, ICDE 2021
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Despite strong theoretical reasons for assuming that abstract representations organize complex action sequences in terms of subplans (chunks) and sequential positions, we lack methods to directly track such content-independent, hierarchical representations in humans. We applied time-resolved, multivariate decoding analysis to the pattern of rhythmic EEG activity that was registered while participants planned and executed individual elements from pre-learned, structured sequences. Across three experiments, the theta and alpha-band activity coded basic elements and abstract control representations, in particular, the ordinal position of basic elements, but also the identity and position of chunks. Further, a robust representation of higher level, chunk identity information was only found in individuals with above-median working memory capacity, potentially providing a neural-level explanation for working-memory differences in sequential performance. Our results suggest that by decoding oscillatory activity we can track how the cognitive system traverses through the states of a hierarchical control structure.more » « less
-
Citations of scientific papers and patents reveal the knowledge flow and usually serve as the metric for evaluating their novelty and impacts in the field. Citation Forecasting thus has various applications in the real world. Existing works on citation forecasting typically exploit the sequential properties of citation events, without exploring the citation network. In this paper, we propose to explore both the citation network and the related citation event sequences which provide valuable information for future citation forecasting. We propose a novel Citation Network and Event Sequence (CINES) Model to encode signals in the citation network and related citation event sequences into various types of embeddings for decoding to the arrivals of future citations. Moreover, we propose a temporal network attention and three alternative designs of bidirectional feature propagation to aggregate the retrospective and prospective aspects of publications in the citation network, coupled with the citation event sequence embeddings learned by a two-level attention mechanism for the citation forecasting. We evaluate our models and baselines on both a U.S. patent dataset and a DBLP dataset. Experimental results show that our models outperform the state-of-the-art methods, i.e., RMTPP, CYAN-RNN, Intensity-RNN, and PC-RNN, reducing the forecasting error by 37.76% - 75.32%.more » « less
-
We present a novel self-supervised approach for hierarchical representation learning and segmentation of perceptual inputs in a streaming fashion. Our research addresses how to semantically group streaming inputs into chunks at various levels of a hierarchy while simultaneously learning, for each chunk, robust global representations throughout the domain. To achieve this, we propose STREAMER, an architecture that is trained layer-by-layer, adapting to the complexity of the input domain. In our approach, each layer is trained with two primary objectives: making accurate predictions into the future and providing necessary information to other levels for achieving the same objective. The event hierarchy is constructed by detecting prediction error peaks at different levels, where a detected boundary triggers a bottom-up information flow. At an event boundary, the encoded representation of inputs at one layer becomes the input to a higher-level layer. Additionally, we design a communication module that facilitates top-down and bottom-up exchange of information during the prediction process. Notably, our model is fully self-supervised and trained in a streaming manner, enabling a single pass on the training data. This means that the model encounters each input only once and does not store the data. We evaluate the performance of our model on the egocentric EPIC-KITCHENS dataset, specifically focusing on temporal event segmentation. Furthermore, we conduct event retrieval experiments using the learned representations to demonstrate the high quality of our video event representations.more » « less
-
Deep learning-based predictive models, leveraging Electronic Health Records (EHR), are receiving increasing attention in healthcare. An effective representation of a patient's EHR should hierarchically encompass both the temporal relationships between historical visits and medical events, and the inherent structural information within these elements. Existing patient representation methods can be roughly categorized into sequential representation and graphical representation. The sequential representation methods focus only on the temporal relationships among longitudinal visits. On the other hand, the graphical representation approaches, while adept at extracting the graph-structured relationships between various medical events, fall short in effectively integrate temporal information. To capture both types of information, we model a patient's EHR as a novel temporal heterogeneous graph. This graph includes historical visits nodes and medical events nodes. It propagates structured information from medical event nodes to visit nodes and utilizes time-aware visit nodes to capture changes in the patient's health status. Furthermore, we introduce a novel temporal graph transformer (TRANS) that integrates temporal edge features, global positional encoding, and local structural encoding into heterogeneous graph convolution, capturing both temporal and structural information. We validate the effectiveness of TRANS through extensive experiments on three real-world datasets. The results show that our proposed approach achieves state-of-the-art performance.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICDE51399.2021.00178
