|
Orolia Presents: GPS World Webinar – Resilient PNT for a 5G World
Panelists will discuss key factors for the successful implementation of 5G technology for 5G infrastructures, automotive, and mission critical applications:
- Testing requirements needed to ensure consistent operations
- Resilient Positioning, Navigation and Timing (PNT) technologies that can help ensure accurate, continuous operations for critical applications during interference or signal loss.
|
|
|
Which is the bigger problem? Risk assessment for PNT with Dana Goward
This webinar is part of the Resilient Positioning Navigation and Timing Seminar Series. Find out more about this series at: https://rin.org.uk/events/EventDetail... Presenter: Dana Goward, President & Director, Resilient Navigation and Timing Foundation
|
|
Assured-Positioning, Navigation and Timing (A-PNT)
YouTube search...
...Google search
Like the GPS units in many automobiles today, a simple receiver and some processing power is all that is needed for accurate navigation. But, what if the GPS satellites suddenly became unavailable due to malfunction, enemy action or simple interference, such as driving into a tunnel? Unavailability of GPS would be inconvenient for drivers on the road, but could be disastrous for military missions. Extreme Miniaturization: Seven Devices, One Chip to Navigate without GPS | US Defense Advanced Research Projects Agency (DARPA)
National Positioning, Navigation, and Timing (PNT) Architecture | U.S. Department of Transportation
|
System Performance and Resilience or: What could possibly go wrong? with Prof Marek Ziebart
Royal Institute of Navigation. Presenter: Prof Marek Ziebart, Director, Space Geodesy and Navigation Group, University College London Website: https://rin.org.uk/
|
|
|
|
|
When GNSS fails, what will you do? - MarRINav!
Royal Institute of Navigation. This webinar, called 'When GNSS fails, what will you do? - MarRINav! ' features presentations and comments form Jonathan Turner (NLA Int.), Dr Alan Grant (GLA), and Dana Goward (RNTF). The webinar provides analysis and insights from Phase 1 of the Maritime Resilience and Integrity of Navigation (MarRINav) project. To download the full transcript of Jonathan's presentation please follow this link: https://rin.org.uk/resource/resmgr/fi... Many thanks to all co-sponsors of this webinar: Resilient Navigation and Timing Foundation, Institute of Navigation, GPS World, The Maritime Executive, and of course the MarRINav project. Website: https://rin.org.uk/
|
|
|
|
The Coming Revolution in MEMS Gyroscopes and MEMS Inertial Sensors
Wireless Integrated MicroSensing & Systems - WIMS2 Relevant for automotive robotic drone wearable applications.
|
|
|
Enabling the Next Generation of GPS Technology with Supercorrelation with Dr Ramsey Faragher
Royal Institute of Navigation. Focal Point Positioning have developed a new method for processing GNSS radio signals called Supercorrelation which dramatically improves the performance of the earliest stage of radio processing. The software update removes multipath interference at the correlator level, and provides the ability to determine signal arrival angle without adding any new hardware to a standard GNSS device. In this webinar Dr Ramsey Faragher, Founder/CEO of FocalPoint will explain how the technology works, will cover some of the challenges that FocalPoint have overcome in deploying it on very low cost platforms, and will show off the new capabilities that it unlocks. Website: https://rin.org.uk/
|
|
|
|
AgilLOC Global Navigation Satellite System (GNSS) Anti-Jamming and Spoofing Capability
AgilLOC Antenna Element Compact (AEC) & Resilient Time Source (RTS) provide assured access to Position, Navigation and Timing (PNT) information for mission-critical systems. AEC provides GNSS anti-jam capability under denied environment through adaptive nulling of interference sources. RTS provides add-on GNSS anti-spoof capability and timing resiliency to ensure the integrity of the GNSS.
|
|
|
GRCon20 - Software defined radio based Global Navigation Satellite System real time spoofing....
Software defined radio based Global Navigation Satellite System real time spoofing detection and cancellation Presented by Jean-Michel Friedt,, D. Rabus and G. Goavec-Merou at GNU Radio Conference 2020 https://gnuradio.org/grcon20 Global Navigation Satellite Systems (GNSS) -- most significantly the Global Positioning System (GPS) -- have become ubiquitous to most daily activities, from positioning and navigation to long range time synchronization or distributed energy production ("smart grid"). While initially developed as a military system hardly accessible to civilians, the advent of Software Defined Radio jamming and spoofing capabilities emphasize the low security of GNSS weak signals emitted from satellites orbiting the Earth 20000 km away. While a properly spoofing signal cannot be detected after a consumer-grade receiver has decoded the radiofrequency signal, addressing at the radiofrequency wave level the signal integrity provides the solution of identifying spoofing with all satellites appearing at the same direction of arrival. This classical beamforming analysis -- Controlled Reception Pattern Antenna (CRPA) with multiple antenna reception and phase analysis -- is demonstrated using commercial, off the shelf software defined radio platform receivers (Ettus Research B210) running the real-time GNSS decoder gnss-sdr based on GNU Radio running on embedded boards such as the Raspberry Pi4.
|
|
|
|
PULP-DroNet -- Autonomous Artificial Intelligence-powered Nano-Drone
PULP-DroNet is a Deep Learning-powered visual navigation engine that enables autonomous navigation of a pocket-size quadrotor in a previously unseen environment.
Thanks to PULP-DroNet the nano-drone can explore the environment, avoiding collisions also with dynamic obstacles, in complete autonomy -- no human operator, no ad-hoc external signals, and no remote laptop! This means that all the complex computations are done directly aboard the vehicle and very fast. The visual navigation engine is composed of both a software and a hardware part. The former is based on the previous DroNet [1] project developed by the RPG [2] from the University of Zürich (UZH). DroNet is a shallow convolutional neural network (CNN) which has been used to control a standard-size quadrotor in a set of environments via remote computation. The hardware soul of PULP-DroNet is embodied by the PULP-Shield an ultra-low power visual navigation module featuring a Parallel Ultra-Low-Power (PULP) GAP8 System-on-Chip (SoC) from GreenWaves Technologies [3], an ultra-low power camera, and off-chip Flash/DRAM memory; the shield is designed as a pluggable PCB for the Crazyflie 2.0 [4] nano-drone. Then, we developed a general methodology for deploying state-of-the-art Deep Learning algorithms on top of ultra-low power embedded computation nodes, like a miniaturized drone. Our novel methodology allowed us first to deploy DroNet on the PULP-Shield, and then demonstrating how it enables the execution the CNN on board the CrazyFlie 2.0 within only 64-284mW and with a throughput of 6-18 frame-per-second! Finally, we field-prove our methodology presenting a closed-loop fully working demonstration of vision-driven autonomous navigation relying only on onboard resources, and within an ultra-low power budget. We release here, as open source, all our code, hardware designs, datasets, and trained networks. Reference: D. Palossi, F. Conti, and L. Benini An Open Source and Open Hardware Deep Learning-powered Visual Navigation Engine for Autonomous Nano-UAVs Preprint: https://arxiv.org/abs/1905.04166 PULP-Platform Project Webpage: https://www.pulp-platform.org/
|
|
|
GPS Spoofing and Jamming: Learn how to protect against threats to GNSS systems
Steatite
|
|
|
|
Securing Positioning & Timing 7: Improving Performance by Augmenting GPS/GNSS
Royal Institute of Navigation. The seventh, and final, of a series of webinars from the Securing Positioning & Timing short course. This webinar covers Improving Performance by Augmenting GPS/GNSS. Presented by Prof Terry Moore. Supported by the UK Space Agency. Website: https://rin.org.uk/
|
|
|
Assess performance & reduce risks across a multisensor, multi-system integration with Peter Rylands
Royal Institute of Navigation. Presenter: Peter Rylands, Product Manager, Oxford Technical Solutions Ltd
|
|
|
|
Resilient PNT for Unmanned Systems
Orolia. Demand for unmanned systems is growing exponentially across defense and civil/commercial organizations, for applications ranging from military missions and intelligence surveillance to border security and precision agriculture.
Resilient Positioning, Navigation and Timing (PNT) data and GPS/GNSS signals are critical for unmanned systems in order to successfully pilot and control aircraft, vehicle and vessel onboard systems.
|
|
Something is jamming GPS over Europe...
Geolocation: Locating GPS/GNSS Jamming and Spoofing
YouTube search...
...Google search
|
Securing Positioning & Timing 3: Detecting and Characterising GPS/GNSS Jamming & Spoofing
Royal Institute of Navigation. The third of a series of webinars from the Securing Positioning & Timing short course. This webinar covers Detecting and Characterising GPS/GNSS Jamming & Spoofing. Presented by Dr Mark Dumville. Supported by the UK Space Agency. Website: https://rin.org.uk/
|
|
|
|
Securing Positioning & Timing 4: Locating GPS/GNSS Jamming and Spoofing
Royal Institute of Navigation. The fourth of a series of webinars from the Securing Positioning & Timing short course. This webinar covers Locating GPS/GNSS Jamming and Spoofing. Presented by Mike Jones. Supported by the UK Space Agency. Website: https://rin.org.uk/
|
|
|
GNSS Jamming - Crowd Sourcing Detection and Geolocation
GNSS Jamming - Crowd Sourcing Detection and Geolocation webinar by InsideGNSS
|
|
|
|
Harris Corporation - Detect and Locate GPS Jamming with Signal Sentry™ 1000
Harris Corporation The Global Positioning System—GPS—is an essential element of the global information infrastructure. GPS jamming devices are becoming cheaper and more accessible, creating a greater need to protect from a diverse range of threats. Harris Signal Sentry 1000 is a GPS interference detection and geolocation solution. It provides a web-based visualization tool to support timely and effective actionable intelligence.
|
|
|
Why isn’t my GPS receiver consistently more accurate? with John Pottle
Royal Institute of Navigation. Presenter: John Pottle, Director of the RIN Website: https://rin.org.uk/
|
|
|
|
Current threats to GNSS: An update of incidents and impacts with Guy Buesnel
Royal Institute of Navigation. Presenter: Guy Buesnel, PNT Security Technologist Spirent Communications plc. Website: https://rin.org.uk/
|
|
Software-defined Global Navigation Satellite Systems (GNSS)
YouTube search...
...Google search
|
Dr. Carles Fernandez: An Open Source Global Navigation Satellite Systems Software-Defined Receiver
Software Defined Radio Academy GNSS-SDR (see https://gnss-sdr.org) is an open source, software-defined Global Navigation Satellite Systems (GNSS) receiver. This software application takes care of all the digital signal processing chain (from the output of the Analog-to-Digital Converter of a radio-frequency front-end, or from raw sam- ples stored in a file), performing signal acquisition and tracking of the available satellite signals, decoding the navigation message and computing the observables needed by positioning algorithms, which ultimately compute the navigation solution. Several outputs are provided in standard formats, including RINEX observation and navigation files, RTCM-104 v3.2 message streaming via TCP/IP and NMEA-0183, as well as KML, GeoJSON, and GPX files for Geographic Information Systems, map representation and Earth browsers. Currently, the software is able to process GLONASS L1 C/A, GPS L1 C/A, Galileo E1b/c, BeiDou B1I, BeiDou B3I, GLONASS L2 C/A, GPS L2C, GPS L5 and Galileo E5a signals, in all possible combinations, including multi-constellation and multi-frequency configurations. The software leverages on the GNU Radio framework, inheriting multithreading scheduling and a modular, scalable architecture. The software is designed to facilitate the inclusion of new signal processing techniques, offering an easy way to measure their impact in the overall receiver performance under fair conditions, as well as the expansion to other signals.
|
|
|
|
GRCon20 - Software defined radio based Global Navigation Satellite System real time spoofing....
Software defined radio based Global Navigation Satellite System real time spoofing detection and cancellation Presented by Jean-Michel Friedt,, D. Rabus and G. Goavec-Merou at GNU Radio Conference 2020 https://gnuradio.org/grcon20 Global Navigation Satellite Systems (GNSS) -- most significantly the Global Positioning System (GPS) -- have become ubiquitous to most daily activities, from positioning and navigation to long range time synchronization or distributed energy production ("smart grid"). While initially developed as a military system hardly accessible to civilians, the advent of Software Defined Radio jamming and spoofing capabilities emphasize the low security of GNSS weak signals emitted from satellites orbiting the Earth 20000 km away. While a properly spoofing signal cannot be detected after a consumer-grade receiver has decoded the radiofrequency signal, addressing at the radiofrequency wave level the signal integrity provides the solution of identifying spoofing with all satellites appearing at the same direction of arrival. This classical beamforming analysis -- Controlled Reception Pattern Antenna (CRPA) with multiple antenna reception and phase analysis -- is demonstrated using commercial, off the shelf software defined radio platform receivers (Ettus Research B210) running the real-time GNSS decoder gnss-sdr based on GNU Radio running on embedded boards such as the Raspberry Pi4.
|
|
Long Range Navigation (LORAN)
YouTube search...
...Google search
|
BBC Click no GPS radio eLoran instead. Filmed at the Port of Felixstowe
BBC Click report at the Port of Felixstowe demonstrating the loss of GPS and using eLoran as an alternative
|
|
Quantum Sensors in Navigation
YouTube search...
...Google search
- Quantum
- UK Research and Innovation
- Review of Quantum Navigation | Donghui Feng - IOP Conference Series: Earth and Environmental Science
- Quantum Sensing Technology Growing Rapidly to Enable Ultra Sensitive Quantum RADARS, Imaging, and Navigation | Rajesh Uppal - International Defence Security & Technology
Typically, the performance of measurement devices is limited by deleterious effects such as thermal noise and vibration. Notable exceptions are atomic clocks, which operate very near their fundamental limits. Driving devices to their physical limits will open new application spaces critical to future DoD systems. Indeed, many defense-critical applications already require exceptionally precise time and frequency standards enabled only by atomic clocks. The Global Positioning System (GPS) and the internet are two key examples. Measurement systems based on atomic physics benefit from the exquisite properties of the atom. Among these are (a) precise frequency transitions, (b) the ability to initialize, control, and readout the atomic state and (c) environmental isolation. In addition, atomic properties are absolute, and do not “drift” over time. In this sense, atoms are self-calibrated, making them ideal for precision sensing. Quantum-Assisted Sensing and Readout (QuASAR) | US Defense Advanced Research Projects Agency (DARPA)
|
Quantum Sensors in Navigation with Roger McKinlay, George Shaw and Kai Bongs
Royal Institute of Navigation Co-hosted by the UK Quantum Technology Hub Sensors and Timing and the Royal Institute of Navigation Presenters: Professor Kai Bongs, Principal Investigator at the UK Quantum Technology Hub Sensors and Timing; Roger McKinlay, Challenge Director for Quantum Technologies, UK Research and Innovation; George Shaw, Principal Systems Engineer, General Lighthouse Authority In this trio of presentations Roger, George and Kai discuss quantum sensors in navigation... Roger covers systems considerations in PNT and vulnerabilities in GNSS, UK industry opportunities and IUK programmes George covers current maritime navigation challenges, the resilient PNT system-of-systems approach and opportunity for Quantum Sensor technology insertion Kai covers quantum Sensor developments towards navigation solutions Website: https://rin.org.uk/
|
|
|
|
Nanoscale Quantum Sensing - Prof. Jorg Wrachtrup
Bar-Ilan University Nanoscale Quantum Sensing - a lecture by Prof. Jorg Wrachtrup of the Institute for Quantum Science and Technology in Germany. This lecture was given during the conference QUEST - Quantum Entanglement Science & Technology, held by Bar-Ilan University's Physics Department in June 2017.
|
|
Spacetime
YouTube search...
...Google search
Spacetime is a mathematical model that fuses the three dimensions of space (length, width, height) and the single dimension of time into a single four-dimensional continuum.
Before the 20th century, scientists viewed space and time as completely separate entities. Space was a static stage where events occurred, and time was a constant, universal clock. Albert Einstein and mathematician Hermann Minkowski revolutionized this view by proving they are inextricably linked.
Spacetime Core Concept: The Fabric of the Universe
In physics, every occurrence is an "event" located at a specific point in spacetime, described by four coordinates:
- <math>(x, y, z, t)</math>
Imagine meeting a friend. To ensure you meet, you must provide a specific location (spatial coordinates) and a specific time (temporal coordinate). If you omit the time, the meeting cannot happen.
General Relativity: Gravity as Curvature
The most famous application of spacetime is in Einstein’s General Theory of Relativity.
- Newton's View: Gravity is an invisible force that pulls objects together.
- Einstein's View: Gravity is not a force; it is the curvature of spacetime.
Massive objects (like the Earth or Sun) warp the fabric of spacetime around them. Smaller objects (like the Moon or a satellite) do not "feel" a force pulling them; they are simply following the straightest possible path (a geodesic) along this curved surface.
Visualizing Spacetime: The Light Cone
Physicists often use Minkowski diagrams to visualize spacetime. In these graphs, time is usually plotted on the vertical axis and space on the horizontal axis.
- World Line: A line representing an object's path through time. Even if an object is stationary in space, it moves through time, creating a straight vertical world line.
- Light Cone: Since nothing can travel faster than light, light spreading out from a single event forms a "cone" shape in the diagram. Events inside this cone can affect one another (causality); events outside are effectively disconnected.
Key Implications
- Time dilation: Because space and time are linked, moving through space affects movement through time. The faster an object travels through space, the slower it moves through time relative to a stationary observer.
- No Universal "Now": There is no single clock for the universe. Two observers moving at different speeds or in different gravitational fields will disagree on when an event happened (Relativity of simultaneity).
Comparison Table
| Concept
|
Classical View (Newton)
|
Spacetime View (Einstein)
|
| Space
|
A static, rigid stage.
|
A flexible fabric that can bend and twist.
|
| Time
|
Universal and constant for everyone.
|
Relative; flows at different rates for different observers.
|
| Gravity
|
A force acting at a distance.
|
The curvature of the geometry of spacetime.
|
|
Time Does Not Exist. Let me explain with a graph.
How do we really move through spacetime? Sadly the books have sold out.
|
|
|
|
What Exactly is Spacetime? Explained in Ridiculously Simple Words
Spacetime, as a concept, is related to a space that consists of 4 dimensions instead of the regular 3-dimensional space. As early as 1905, Einstein proposed a now widely popular theory that the speed of light is independent of the motion of all observers, and that space and time are interconnected in a single continuum. This theory, which is now a cornerstone of modern and quantum physics, is known as Einstein’s special theory of relativity. Einstein's proposed idea of a single continuum where space and time are interwoven is what people call “space-time”.
According to this theory, time—which has traditionally been considered an independent entity according to the principles of classical physics—is affected when a body moves through space. This happens because, according to the theory, time and space are connected and part of a single continuum—spacetime.
In this video, we discuss spacetime in absolutely simple words: what exactly is spacetime and how is it related to the force of gravitation and Einstein’s theory of relativity?
|
|
|
This Is Why Time Might Not Actually Exist
Quantum Entanglement May Reveal a Reality We Can't Handle
|
|
|
|
Longitude
YouTube search...
...Google search
|
Longitude
Dava Sobel's book is a captivating story, such a good film with an impressive cast. What a genius Mr Harrison was. Such determination. A lesson to us all..
|
|
|
|
The Clock That Changed the World (BBC History of the World)
SciShow It’s time for another leap second! Join SciShow as we celebrate by exploring the long and strange history of timekeeping. Hosted by: Michael Aranda Dooblydoo thanks go to the following Patreon supporters -- we couldn't make SciShow without them! Shout out to Justin Ove, Justin Lentz, David Campos, John Szymakowski, Peso255, Jeremy Peng, Avi Yaschin, and Fatima Iqbal. Like SciShow? Want to help support us, and also get things to put on your walls, cover your torso and hold your liquids? Check out our awesome products over at DFTBA Records: https://dftba.com/scishow Or help support us by becoming our patron on Patreon: https://www.patreon.com/scishow
|
|
Natural Navigation
YouTube search...
...Google search
|
Henrik Mouritsen on How do migratory birds find their way?
Prof. Henrik Mouritsen (University of Oldenburg, Germany) on "How do migratory birds find their way?", at the FENS Hertie Winter School 2017 on Neural Control of Behaviour - Series 1: Navigation, 10-16 December 2017, Obergurgl, Austria.
|
|
|
|
Animal Navigation || Radcliffe Institute
Lizabeth Cohen 00:16 Dean, Radcliffe Institute for Advanced Study Howard Mumford Jones Professor of American Studies Department of History Harvard University John Huth 7:32 Donner Professor of Science
Codirector of the Science Program Radcliffe Institute for Advanced Study Harvard University ANIMAL NAVIGATION 16:50 Susanne Åkesson 19:53 Professor and Principal Investigator Centre for Animal Movement Research, Lund University (Sweden) Introduced by Scott Edwards 17:51 Professor of Organismic and Evolutionary Biology Curator of Ornithology Alexander Agassiz Professor of Zoology in the Museum of Comparative Zoology Harvard University
|
|
Molecular Clock
Scientists use the molecular clock, which assumes steady genetic changes, to estimate species divergence times, but recent models like the Covariant Evolutionary Tempo (CET) by Budd & Mann suggest evolution isn't always steady, predicting rapid bursts in major groups (like mammals or birds) early on, explaining mismatches with fossil records by showing faster initial evolution and diversification, thus refining our understanding of how large animal groups rapidly emerge.
How the Molecular Clock Works
- Rate of Mutation: The core idea is that mutations in DNA accumulate at a relatively constant rate over time.
- Genetic Differences: By comparing DNA or protein sequences between species, scientists count the genetic differences.
- Dating Divergence: More differences imply a longer time since the species shared a common ancestor, allowing estimation of evolutionary timelines.
Challenges & New Models (Budd & Mann's CET)
the Covariant Evolutionary Tempo model suggests that when a big group of organisms appear, evolution actually speeds up. This would make it appear like more time was passing when evolution was really on fast-forward, differentiating into various groups that eventually appeared in the fossil record. “While the speeding clock idea needs testing,” Telford wrote, “it could explain other mismatches between molecular clocks and the fossil record.”
- Fossil Record Mismatch: Molecular clocks sometimes suggest earlier origins for animal groups than the fossil record shows, creating a gap (e.g., the Cambrian explosion).
- The CET Model: This model proposes that when a major group starts to diversify, it experiences:
- Explosive Radiation: Rapid increases in species diversity.
- Elevated Molecular Rates: Faster rates of genetic change.
- Impact: This explains why fossil records show sudden appearances of major groups, as they truly did evolve and diversify quickly, rather than gradually over long periods, says this article from Uppsala University.
Significance
- Refined Timelines: The new models provide a more nuanced understanding of evolutionary history, reconciling molecular data with fossil evidence.
- Understanding Major Events: Helps explain rapid evolutionary events, like the emergence of mammals after dinosaur extinction, where a surviving lineage exploded in diversity.
- Connecting Disciplines: Bridges gaps between molecular biology, paleontology, and geology to build more accurate evolutionary trees.
Time Travel in Fiction
Time travel in science and fiction serves as a rich narrative tool to explore causality, free will, and the malleability of timelines. Scientific theories like wormholes, relativistic travel involving time dilation, and closed timelike curves provide plausible mechanisms explanation of how time travel functions. In different popular movies, books, & shows – not how it works “under the hood", but how it causally affects the perspective of characters’ timelines (who has free will? can you change things by going back to the past or forwards into the future?). In particular, I explain Ender's Game, Planet of the Apes, Harry Potter and the Prisoner of Azkaban, Primer, Bill & Ted’s Excellent Adventure, Back to the Future, Groundhog Day, Looper, the video game “Braid”, and Lifeline. Whether driven by science or narrative needs, these portrayals reflect how characters experience and manipulate time, raising questions about fate, agency, and the consequences of tampering with the past
Time & Music
YouTube search...
...Google search
Sequence/Time-based Algorithms
Time-based AI algorithms are algorithms that use time series data to make predictions or analyses. Time series data are data that are collected over time and have a temporal order. For example, the daily temperature, the stock prices, or the number of visitors to a website are all time series data. These algorithms can be used for a variety of purposes, such as forecasting future values, detecting trends and patterns, and making informed decisions based on historical data. They can be applied to many different fields, including finance, economics, meteorology, and healthcare.
Whenever we have developed better clocks, we’ve learned something new about the world. - Alexander Smith New Time Dilation Phenomenon Revealed: Timekeeping Theory Combines Quantum Clocks and Einstein’s Relativity - Dartmouth College
Common
There are different types of sequence/time-based AI algorithms, depending on the goal and the method of the algorithm. Some of the most common ones are:
- Time Series Forecasting:
- Statistical:
- Autoregressive (AR): uses past values of the time series to predict future values. It assumes that the current value is a linear function of previous values. For example, AR can be used to forecast the weather based on historical data.
- Autoregressive Integrated Moving Average (ARIMA): is an extension of AR that also accounts for the trend and the seasonality of the time series. It uses differencing to make the time series stationary (i.e., having constant mean and variance) and then applies AR and moving average (MA) models. For example, ARIMA can be used to forecast the sales of a product based on past sales and seasonal patterns.
- Seasonal Autoregressive Integrated Moving Average (SARIMA): is a further extension of ARIMA that also accounts for the cyclic variations of the time series. It uses seasonal differencing and seasonal AR and MA models to capture the periodic fluctuations of the time series. For example, SARIMA can be used to forecast the electricity demand based on past demand and seasonal factors.
- Exponential Smoothing (ES): uses weighted averages of past values of the time series to predict future values. It gives more weight to recent values than older values, and it can also incorporate trend and seasonality components. For example, ES can be used to forecast the inventory level based on past demand and supply.
- Deep Learning:
- Prophet: is a modern and flexible approach to time series forecasting developed by Facebook. It uses a decomposable model that consists of trend, seasonality, and holiday components, and it allows for adding custom effects and prior information. For example, Prophet can be used to forecast the web traffic for a data science blog website based on past traffic and special events.
- Neural Turing Machine (NTM): the fuzzy pattern matching capabilities of Neural Networks with the algorithmic power of programmable computers. NTMs are an instance of Memory Augmented Neural Networks, a new class of Recurrent Neural Network (RNN)s which decouple computation from memory by introducing an external memory unit. NTMs have demonstrated superior performance over Long Short-Term Memory Cells in several sequence learning tasks.
- Neural Networks:
- Recurrent Neural Network (RNN): is a type of Deep Learning model that can process sequential data such as time series. It uses a network of neurons that have feedback loops, which enable them to store information from previous inputs. For example, RNN can be used to forecast the prices of Bitcoin based on past prices and other factors.
- Gated Recurrent Unit (GRU): are a gating mechanism in Recurrent Neural Network (RNN) architecture. Like other RNNs, a GRU can process sequential data such as time series, natural language, and speech1. The GRU is similar to a Long Short-Term Memory (LSTM) with a forget gate, but has fewer parameters than LSTM, as it lacks an output gate. This means that GRUs are generally easier and faster to train than their LSTM counterparts. GRUs have been found to perform similarly to LSTMs on certain tasks such as polyphonic music modeling, speech signal modeling, and natural language processing. They have shown that gating is indeed helpful in general.
- Long Short-Term Memory (LSTM): is a special type of RNN that can handle long-term dependencies in sequential data. It uses a memory cell that can store, update, and forget information over time, and it has gates that control the flow of information in and out of the cell. For example, LSTM can be used to forecast the generation of wind power based on past generation and weather conditions:
- Bidirectional Long Short-Term Memory (BI-LSTM): is a type of Recurrent Neural Network (RNN) architecture that processes data in both forward and backward directions. It consists of two LSTMs: one taking the input in a forward direction, and the other in a backward direction. BI-LSTMs effectively increase the amount of information available to the network, improving the context available to the algorithm. For example, knowing what words immediately follow and precede a word in a sentence. Compared to LSTM, BI-LSTM combines the forward hidden layer and the backward hidden layer, which can access both the preceding and succeeding contexts¹. This feature of flow of data in both directions makes the BI-LSTM different from other LSTMs. BI-LSTMs have been successfully applied to various tasks such as natural language processing, speech recognition, and traffic forecasting.
- Bidirectional Long Short-Term Memory (BI-LSTM) with Attention Mechanism: is a type of Recurrent Neural Network (RNN) architecture that processes data in both forward and backward directions, and uses an attention mechanism to weigh the importance of different parts of the input sequence. The attention mechanism allows the network to focus on specific parts of the input sequence when making predictions, rather than treating all parts of the sequence equally. This can be particularly useful when dealing with long input sequences, where some parts of the sequence may be more relevant to the prediction than others. BI-LSTMs with Attention Mechanism have been successfully applied to various tasks such as text classification, Sentiment Analysis, and human activity recognition.
- Average-Stochastic Gradient Descent (SGD) Weight-Dropped LSTM (AWD-LSTM): is a variant of LSTM that employs DropConnect for regularization, as well as NT-ASGD for optimization. NT-ASGD stands for non-monotonically triggered averaged stochastic gradient descent, which returns an average of the last iterations of weights. AWD-LSTM has shown great results on both word-level and character-level models. It has been used in research papers on word-level models and has shown great results on character-level models as well.
- Sequence to Sequence (Seq2Seq): can map a variable-length input sequence to a variable-length output sequence. It is often used for natural language processing tasks, such as machine translation, text summarization, conversational models, and question answering. The Seq2Seq algorithm consists of two main components: an encoder and a decoder. The encoder reads the input sequence one timestep at a time and produces a hidden vector representation of the input. The decoder then uses the hidden vector as the initial state and generates the output sequence one timestep at a time, using the previous output as the input context.
- Transformer: is a state-of-the-art Deep Learning model that can process sequential data such as time series. It uses layers of attention mechanisms that can learn how to focus on relevant parts of the input data, and it can handle long-term dependencies and parallel computations efficiently. For example, Transformer can be used to forecast the spread of COVID-19 based on past cases and interventions. Transformer can process sequential data using layers of attention mechanisms, without using recurrent or convolutional layers. It can handle long-term dependencies and parallel computations efficiently, and it can achieve better results than RNN-based Seq2Seq models on various tasks.
- Generative Pre-trained Transformer (GPT): are a family of language models that use Deep Learning techniques to generate natural language text. They are based on the transformer architecture and can be fine-tuned for various natural language processing tasks such as text generation, language translation, and text classification. The first GPT was introduced in 2018 by the American artificial intelligence (AI) company OpenAI. GPT models are artificial Neural Networks that are based on the transformer architecture, pre-trained on large data sets of unlabelled text, and able to generate novel human-like content
- Attention Mechanism: allows the decoder to selectively focus on different parts of the input sequence when generating the output, instead of relying on a single fixed vector. This can improve the performance and accuracy of the Seq2Seq model, especially for long sequences
- Transformer-XL: is a transformer-based language model that introduces the notion of recurrence to the deep self-attention network. It was designed to enable learning dependency beyond a fixed length without disrupting temporal coherence. The model consists of a segment-level recurrence mechanism and a novel positional encoding scheme. This method not only enables capturing longer-term dependency, but also resolves the context fragmentation problem. As a result, Transformer-XL learns dependency that is 80% longer than RNNs and 450% longer than vanilla Transformers, achieves better performance on both short and long sequences, and is up to 1,800+ times faster than vanilla Transformers during evaluation.
- Beam search: is a technique to find the most probable output sequence given the input sequence, by keeping track of multiple candidate sequences and expanding them based on their probabilities. This can improve the quality and diversity of the output, compared to using a greedy or random search.
- Convolutional Neural Network (CNN): is another type of Deep Learning model that can process sequential data such as time series. It uses layers of filters that can extract features from local regions of the input data, and it can capture complex patterns and relationships in the data. For example, CNN can be used to forecast an avalanche in a famous ski resort based on past snowfall and temperature data.
- Spatial-Temporal Dynamic Network (STDN): a Deep Learning framework proposed to address the challenge of modeling complex spatial dependencies and temporal dynamics in traffic prediction. A flow gating mechanism is introduced to learn the dynamic similarity between locations, and a periodically shifted attention mechanism is designed to handle long-term periodic temporal shifting. This approach has been shown to be effective in predicting taxi demand
- Other:
- Gaussian Process (GP): is a type of probabilistic model that can handle uncertainty and noise in time series data. It uses a function that defines how similar any two points in the input space are, and it produces a distribution over possible outputs for any given input. For example, GP can be used to forecast the depletion level of stocks in stores based on past sales and inventory data.
- End-to-End Speech: translation is an approach to speech translation that is gaining high interest from the research world in the last few years. It consists of using a single Deep Learning model that learns to generate translated text of the input audio in an end-to-end fashion. This approach, known as “end-to-end” or “direct” ST, supposes many advantages over the former, such as avoiding the concatenation of errors, the direct use of prosodic from speech and a lower inference time.
- (Tree) Recursive Neural (Tensor) Network (RNTN): type of Neural Network that is mostly used for natural language processing. It has a tree structure with a neural net at each node. The purpose of these nets is to analyze data that have a hierarchy of structure. An RNTN is a powerful tool for deciphering and labeling patterns. Structurally, an RNTN is a binary tree with three nodes: a root and two leaves. The root and leaf nodes are not neurons, but instead, they are groups of neurons – the more complicated the input data, the more neurons are required. RNTNs have been successfully applied to Sentiment Analysis, where the input is a sentence in its parse tree structure, and the output is the classification for the input sentence, i.e., whether the meaning is very negative, negative, neutral, positive, or very positive
- Temporal Difference (TD) Learning: refers to a class of model-free Reinforcement Learning (RL) methods which learn by bootstrapping from the current estimate of the value function. These methods sample from the environment, like Monte Carlo methods, and perform updates based on current estimates, like dynamic programming methods. While Monte Carlo methods only adjust their estimates once the final outcome is known, TD methods adjust predictions to match later, more accurate, predictions about the future before the final outcome is known.
Time is an indispensable concept within computer programs and applications. Without this concept, we would not be able to access any transport layer security (TLS) based websites, exchange data, or utilize various cryptographic algorithms. | Olga Hryniuk - Input Output
|