The Importance of Loss Functions for Increasing the Generalization Abilities of a Deep Learning-Based Next Frame Prediction Model for Traffic Scenes

Aigner, Sandra; Körner, Marco

doi:10.3390/make2020006

Cited by 5 publications

(1 citation statement)

References 24 publications

(39 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This makes difficult to draw predictions maintaining the consistency with our visual similarity notion. A recent study [39] performed an in-depth analysis of the generalization capabilities of different loss functions for the video prediction task. Besides video prediciton, the impact of different loss functions was analyzed in image restoration restoration [40], classification [41], camera pose regression [42] and structured prediction [43], among others.…”

Section: The Devil Is In the Loss Functionmentioning

confidence: 99%

A Review on Deep Learning Techniques for Video Prediction

Oprea

Martinez-Gonzalez

García-García

et al. 2022

IEEE Trans. Pattern Anal. Mach. Intell.

137

View full text Add to dashboard Cite

The ability to predict, anticipate and reason about future outcomes is a key component of intelligent decision-making systems. In light of the success of deep learning in computer vision, deep-learning-based video prediction emerged as a promising research direction. Defined as a self-supervised learning task, video prediction represents a suitable framework for representation learning, as it demonstrated potential capabilities for extracting meaningful representations of the underlying patterns in natural videos. Motivated by the increasing interest in this task, we provide a review on the deep learning methods for prediction in video sequences. We firstly define the video prediction fundamentals, as well as mandatory background concepts and the most used datasets. Next, we carefully analyze existing video prediction models organized according to a proposed taxonomy, highlighting their contributions and their significance in the field. The summary of the datasets and methods is accompanied with experimental results that facilitate the assessment of the state of the art on a quantitative basis. The paper is summarized by drawing some general conclusions, identifying open research challenges and by pointing out future research directions.

show abstract

Section: The Devil Is In the Loss Functionmentioning

confidence: 99%

A Review on Deep Learning Techniques for Video Prediction

Oprea

Martinez-Gonzalez

García-García

et al. 2022

IEEE Trans. Pattern Anal. Mach. Intell.

137

View full text Add to dashboard Cite

show abstract

A Hybrid Artificial Neural Network to Estimate Soil Moisture Using SWAT+ and SMAP Data

Breen

James

White

et al. 2020

MAKE

View full text Add to dashboard Cite

In this work, we developed a data-driven framework to predict near-surface (0–5 cm) soil moisture (SM) by mapping inputs from the Soil & Water Assessment Tool to SM time series from NASA’s Soil Moisture Active Passive (SMAP) satellite for the period 1 January 2016–31 December 2018. We developed a hybrid artificial neural network (ANN) combining long short-term memory and multilayer perceptron networks that were used to simultaneously incorporate dynamic weather and static spatial data into the training algorithm, respectively. We evaluated the generalizability of the hybrid ANN using training datasets comprising several watersheds with different environmental conditions, examined the effects of standard and physics-guided loss functions, and experimented with feature augmentation. Our model could estimate SM on par with the accuracy of SMAP. We demonstrated that the most critical learning of the physical processes governing SM variability was learned from meteorological time series, and that additional physical context supported model performance when test data were not fully encapsulated by the variability of the training data. Additionally, we found that when forecasting SM based on trends learned during the earlier training period, the models appreciated seasonal trends.

show abstract

Application of Machine Learning in Predicting Cement Properties

Shamsan,

Jimenez,

Jandhyala

2024

Day 3 Thu, February 29, 2024

View full text Add to dashboard Cite

Cementing operations consists of placing cement slurries between casing-casing and casing-formation to achieve multiple objectives including zonal isolation, corrosion resistance, isolation of water bearing formations, etc. Cement slurries are tailored in the laboratory environment simulating downhole conditions to achieve the required placement times (i.e. thickening time) for the slurry to remain in a liquid state during pumping. This study presents an artificial intelligence model developed to predict the cement slurries thickening time based on historical experimental data, along with the ability of the model to continue learning based on the flow of output results. Collected data sets of different input parameters were gathered and studied to estimate cement thickening time. Based on statistical analysis, only five inputs (Pressure (psi), measured depth (ft),bottom hole circulating temperature (°F), retarder additive concentration (%), and slurry density (ppg)) had a positive impact on the correlation coefficients between the inputs and target properties. A model was developed using Python programming language to predict the thickening time. The data was fed into the model to train and test the input factors impacting the target property using a neural network method. The model expressed an integrated relationship between the training and validation data which resulted in an optimum learning performance revealed by a loss function plot. As a result, the scatter plot for prediction vs actual has illustrated the prediction of the thickening time with an average accuracy of 83% confirmed by the model box plot. The model validation consisted of employing a set of data to predict the thickening time for more than 11 slurries with an average absolute error of 6%. Model implementation indicated that over 200 hours per month were saved due to reduced number of thickening time test iterations required to be run before executing the job which positively impacting operational efficiencies associated with cost, resource allocation, and investigation assessment. The outcome results from this model and the process used herein could be scaled up to consider other properties, i.e. ultra-sonic cement analyzer UCA, fluid loss, free fluid, etc.

show abstract

The Importance of Loss Functions for Increasing the Generalization Abilities of a Deep Learning-Based Next Frame Prediction Model for Traffic Scenes

Cited by 5 publications

References 24 publications

A Review on Deep Learning Techniques for Video Prediction

A Review on Deep Learning Techniques for Video Prediction

A Hybrid Artificial Neural Network to Estimate Soil Moisture Using SWAT+ and SMAP Data

Application of Machine Learning in Predicting Cement Properties

Contact Info

Product

Resources

About