-
Collage: Light-Weight Low-Precision Strategy for LLM Training
Authors:
Tao Yu,
Gaurav Gupta,
Karthick Gopalswamy,
Amith Mamidala,
Hao Zhou,
Jeffrey Huynh,
Youngsuk Park,
Ron Diamant,
Anoop Deoras,
Luke Huan
Abstract:
Large models training is plagued by the intense compute cost and limited hardware memory. A practical solution is low-precision representation but is troubled by loss in numerical accuracy and unstable training rendering the model less useful. We argue that low-precision floating points can perform well provided the error is properly compensated at the critical locations in the training process. W…
▽ More
Large models training is plagued by the intense compute cost and limited hardware memory. A practical solution is low-precision representation but is troubled by loss in numerical accuracy and unstable training rendering the model less useful. We argue that low-precision floating points can perform well provided the error is properly compensated at the critical locations in the training process. We propose Collage which utilizes multi-component float representation in low-precision to accurately perform operations with numerical errors accounted. To understand the impact of imprecision to training, we propose a simple and novel metric which tracks the lost information during training as well as differentiates various precision strategies. Our method works with commonly used low-precision such as half-precision ($16$-bit floating points) and can be naturally extended to work with even lower precision such as $8$-bit. Experimental results show that pre-training using Collage removes the requirement of using $32$-bit floating-point copies of the model and attains similar/better training performance compared to $(16, 32)$-bit mixed-precision strategy, with up to $3.7\times$ speedup and $\sim 15\%$ to $23\%$ less memory usage in practice.
△ Less
Submitted 6 May, 2024;
originally announced May 2024.
-
Multiple Mobile Target Detection and Tracking in Active Sonar Array Using a Track-Before-Detect Approach
Authors:
Avi Abu,
Nikola Miskovic,
Oleg Chebotar,
Neven Cukrov,
Roee Diamant
Abstract:
We present an algorithm for detecting and tracking underwater mobile objects using active acoustic transmission of broadband chirp signals whose reflections are received by a hydrophone array. The method overcomes the problem of high false alarm rate by applying a track-before-detect approach to the sequence of received reflections. A 2D time-space matrix is created for the reverberations received…
▽ More
We present an algorithm for detecting and tracking underwater mobile objects using active acoustic transmission of broadband chirp signals whose reflections are received by a hydrophone array. The method overcomes the problem of high false alarm rate by applying a track-before-detect approach to the sequence of received reflections. A 2D time-space matrix is created for the reverberations received from each transmitted probe signal by performing delay and sum beamforming and pulse compression. The result is filtered by a 2D constant false alarm rate (CFAR) detector to identify reflection patterns corresponding to potential targets. Closely spaced signals for multiple probe transmissions are combined into blobs to avoid multiple detections of a single object. A track-before-detect method using a Nearly Constant Velocity (NCV) model is employed to track multiple objects. The position and velocity is estimated by the debiased converted measurement Kalman filter. Results are analyzed for simulated scenarios and for experiments at sea, where GPS tagged gilt-head seabream fish were tracked. Compared to two benchmark schemes, the results show a favorable track continuity and accuracy that is robust to the choice of detection threshold.
△ Less
Submitted 16 April, 2024;
originally announced April 2024.
-
Review of Cetacean's click detection algorithms
Authors:
Mak Gracic,
Guy Gubnisky,
Roee Diamant
Abstract:
The detection of echolocation clicks is key in understanding the intricate behaviors of cetaceans and monitoring their populations. Cetacean species relying on clicks for navigation, foraging and even communications are sperm whales (Physeter macrocephalus) and a variety of dolphin groups. Echolocation clicks are wideband signals of short duration that are often emitted in sequences of varying int…
▽ More
The detection of echolocation clicks is key in understanding the intricate behaviors of cetaceans and monitoring their populations. Cetacean species relying on clicks for navigation, foraging and even communications are sperm whales (Physeter macrocephalus) and a variety of dolphin groups. Echolocation clicks are wideband signals of short duration that are often emitted in sequences of varying inter-click-intervals. While datasets and models for clicks exist, the detection and classification of clicks present a significant challenge, mostly due to the diversity of clicks' structures, overlap** signals from simultaneously emitting animals, and the abundance of noise transients from, for example, snap** shrimps and ship** cavitation noise. This paper provides a survey of the many detection and classification methodologies of clicks, ranging from 2002 to 2023. We divide the surveyed techniques into categories by their methodology. Specifically, feature analysis (e.g., phase, ICI and duration), frequency content, energy based detection, supervised and unsupervised machine learning, template matching and adaptive detection approaches. Also surveyed are open access platforms for click detections, and databases openly available for testing. Details of the method applied for each paper are given along with advantages and limitations, and for each category we analyze the remaining challenges. The paper also includes a performance comparison for several schemes over a shared database. Finally, we provide tables summarizing the existing detection schemes in terms of challenges address, methods, detection and classification tools applied, features used and applications.
△ Less
Submitted 7 February, 2024;
originally announced February 2024.
-
Detecting the presence of sperm whales echolocation clicks in noisy environments
Authors:
Guy Gubnitsky,
Roee Diamant
Abstract:
Sperm whales (Physeter macrocephalus) navigate underwater with a series of impulsive, click-like sounds known as echolocation clicks. These clicks are characterized by a multipulse structure (MPS) that serves as a distinctive pattern. In this work, we use the stability of the MPS as a detection metric for recognizing and classifying the presence of clicks in noisy environments. To distinguish betw…
▽ More
Sperm whales (Physeter macrocephalus) navigate underwater with a series of impulsive, click-like sounds known as echolocation clicks. These clicks are characterized by a multipulse structure (MPS) that serves as a distinctive pattern. In this work, we use the stability of the MPS as a detection metric for recognizing and classifying the presence of clicks in noisy environments. To distinguish between noise transients and to handle simultaneous emissions from multiple sperm whales, our approach clusters a time series of MPS measures while removing potential clicks that do not fulfil the limits of inter-click interval, duration and spectrum. As a result, our approach can handle high noise transients and low signal-to-noise ratio. The performance of our detection approach is examined using three datasets: seven months of recordings from the Mediterranean Sea containing manually verified ambient noise; several days of manually labelled data collected from the Dominica Island containing approximately 40,000 clicks from multiple sperm whales; and a dataset from the Bahamas containing 1,203 labelled clicks from a single sperm whale. Comparing with the results of two benchmark detectors, a better trade-off between precision and recall is observed as well as a significant reduction in false detection rates, especially in noisy environments. To ensure reproducibility, we provide our database of labelled clicks along with our implementation code.
△ Less
Submitted 31 December, 2023;
originally announced January 2024.
-
Underwater object classification combining SAS and transferred optical-to-SAS Imagery
Authors:
Avi Abu,
Roee Diamant
Abstract:
Combining synthetic aperture sonar (SAS) imagery with optical images for underwater object classification has the potential to overcome challenges such as water clarity, the stability of the optical image analysis platform, and strong reflections from the seabed for sonar-based classification. In this work, we propose this type of multi-modal combination to discriminate between man-made targets an…
▽ More
Combining synthetic aperture sonar (SAS) imagery with optical images for underwater object classification has the potential to overcome challenges such as water clarity, the stability of the optical image analysis platform, and strong reflections from the seabed for sonar-based classification. In this work, we propose this type of multi-modal combination to discriminate between man-made targets and objects such as rocks or litter. We offer a novel classification algorithm that overcomes the problem of intensity and object formation differences between the two modalities. To this end, we develop a novel set of geometrical shape descriptors that takes into account the geometrical relation between the objects shadow and highlight. Results from 7,052 pairs of SAS and optical images collected during several sea experiments show improved classification performance compared to the state-of-the-art for better discrimination between different types of underwater objects. For reproducibility, we share our database.
△ Less
Submitted 24 April, 2023;
originally announced April 2023.
-
An Efficient Drifters Deployment Strategy to Evaluate Water Current Velocity Fields
Authors:
Murad Tukan,
Eli Biton,
Roee Diamant
Abstract:
Water current prediction is essential for understanding ecosystems, and to shed light on the role of the ocean in the global climate context. Solutions vary from physical modeling, and long-term observations, to short-term measurements. In this paper, we consider a common approach for water current prediction that uses Lagrangian floaters for water current prediction by interpolating the trajector…
▽ More
Water current prediction is essential for understanding ecosystems, and to shed light on the role of the ocean in the global climate context. Solutions vary from physical modeling, and long-term observations, to short-term measurements. In this paper, we consider a common approach for water current prediction that uses Lagrangian floaters for water current prediction by interpolating the trajectory of the elements to reflect the velocity field. Here, an important aspect that has not been addressed before is where to initially deploy the drifting elements such that the acquired velocity field would efficiently represent the water current. To that end, we use a clustering approach that relies on a physical model of the velocity field. Our method segments the modeled map and determines the deployment locations as those that will lead the floaters to 'visit' the center of the different segments. This way, we validate that the area covered by the floaters will capture the in-homogeneously in the velocity field. Exploration over a dataset of velocity field maps that span over a year demonstrates the applicability of our approach, and shows a considerable improvement over the common approach of uniformly randomly choosing the initial deployment sites. Finally, our implementation code can be found in [1].
△ Less
Submitted 10 January, 2023;
originally announced January 2023.
-
Automated Detection of Dolphin Whistles with Convolutional Networks and Transfer Learning
Authors:
Burla Nur Korkmaz,
Roee Diamant,
Gil Danino,
Alberto Testolin
Abstract:
Effective conservation of maritime environments and wildlife management of endangered species require the implementation of efficient, accurate and scalable solutions for environmental monitoring. Ecoacoustics offers the advantages of non-invasive, long-duration sampling of environmental sounds and has the potential to become the reference tool for biodiversity surveying. However, the analysis and…
▽ More
Effective conservation of maritime environments and wildlife management of endangered species require the implementation of efficient, accurate and scalable solutions for environmental monitoring. Ecoacoustics offers the advantages of non-invasive, long-duration sampling of environmental sounds and has the potential to become the reference tool for biodiversity surveying. However, the analysis and interpretation of acoustic data is a time-consuming process that often requires a great amount of human supervision. This issue might be tackled by exploiting modern techniques for automatic audio signal analysis, which have recently achieved impressive performance thanks to the advances in deep learning research. In this paper we show that convolutional neural networks can indeed significantly outperform traditional automatic methods in a challenging detection task: identification of dolphin whistles from underwater audio recordings. The proposed system can detect signals even in the presence of ambient noise, at the same time consistently reducing the likelihood of producing false positives and false negatives. Our results further support the adoption of artificial intelligence technology to improve the automatic monitoring of marine ecosystems.
△ Less
Submitted 28 November, 2022;
originally announced November 2022.
-
Cetacean Translation Initiative: a roadmap to deciphering the communication of sperm whales
Authors:
Jacob Andreas,
Gašper Beguš,
Michael M. Bronstein,
Roee Diamant,
Denley Delaney,
Shane Gero,
Shafi Goldwasser,
David F. Gruber,
Sarah de Haas,
Peter Malkin,
Roger Payne,
Giovanni Petri,
Daniela Rus,
Pratyusha Sharma,
Dan Tchernov,
Pernille Tønnesen,
Antonio Torralba,
Daniel Vogt,
Robert J. Wood
Abstract:
The past decade has witnessed a groundbreaking rise of machine learning for human language analysis, with current methods capable of automatically accurately recovering various aspects of syntax and semantics - including sentence structure and grounded word meaning - from large data collections. Recent research showed the promise of such tools for analyzing acoustic communication in nonhuman speci…
▽ More
The past decade has witnessed a groundbreaking rise of machine learning for human language analysis, with current methods capable of automatically accurately recovering various aspects of syntax and semantics - including sentence structure and grounded word meaning - from large data collections. Recent research showed the promise of such tools for analyzing acoustic communication in nonhuman species. We posit that machine learning will be the cornerstone of future collection, processing, and analysis of multimodal streams of data in animal communication studies, including bioacoustic, behavioral, biological, and environmental data. Cetaceans are unique non-human model species as they possess sophisticated acoustic communications, but utilize a very different encoding system that evolved in an aquatic rather than terrestrial medium. Sperm whales, in particular, with their highly-developed neuroanatomical features, cognitive abilities, social structures, and discrete click-based encoding make for an excellent starting point for advanced machine learning tools that can be applied to other animals in the future. This paper details a roadmap toward this goal based on currently existing technology and multidisciplinary scientific community effort. We outline the key elements required for the collection and processing of massive bioacoustic data of sperm whales, detecting their basic communication units and language-like higher-level structures, and validating these models through interactive playback experiments. The technological capabilities developed by such an undertaking are likely to yield cross-applications and advancements in broader communities investigating non-human communication and animal behavioral research.
△ Less
Submitted 17 April, 2021;
originally announced April 2021.
-
Cooperative Authentication in Underwater Acoustic Sensor Networks
Authors:
Roee Diamant,
Paolo Casari,
Stefano Tomasin
Abstract:
With the growing use of underwater acoustic communications (UWAC) for both industrial and military operations, there is a need to ensure communication security. A particular challenge is represented by underwater acoustic networks (UWANs), which are often left unattended over long periods of time. Currently, due to physical and performance limitations, UWAC packets rarely include encryption, leavi…
▽ More
With the growing use of underwater acoustic communications (UWAC) for both industrial and military operations, there is a need to ensure communication security. A particular challenge is represented by underwater acoustic networks (UWANs), which are often left unattended over long periods of time. Currently, due to physical and performance limitations, UWAC packets rarely include encryption, leaving the UWAN exposed to external attacks faking legitimate messages. In this paper, we propose a new algorithm for message authentication in a UWAN setting. We begin by observing that, due to the strong spatial dependency of the underwater acoustic channel, an attacker can attempt to mimic the channel associated with the legitimate transmitter only for a small set of receivers, typically just for a single one. Taking this into account, our scheme relies on trusted nodes that independently help a sink node in the authentication process. For each incoming packet, the sink fuses beliefs evaluated by the trusted nodes to reach an authentication decision. These beliefs are based on estimated statistical channel parameters, chosen to be the most sensitive to the transmitter-receiver displacement. Our simulation results show accurate identification of an attacker's packet. We also report results from a sea experiment demonstrating the effectiveness of our approach.
△ Less
Submitted 2 January, 2019; v1 submitted 7 June, 2018;
originally announced June 2018.
-
Fair and Throughput-Optimal Routing in Multi-Modal Underwater Networks
Authors:
Roee Diamant,
Paolo Casari,
Filippo Campagnaro,
Oleksiy Kebkal,
Veronika Kebkal,
Michele Zorzi
Abstract:
While acoustic communications have been considered the prominent technology to communicate under water for several years, other technologies are being developed based, e.g., on optical and radio-frequency electro-magnetic waves. Each technology has its own advantages and drawbacks: for example, acoustic signals achieve long communication ranges at order-of-kbit/s bit rate, whereas optical signals…
▽ More
While acoustic communications have been considered the prominent technology to communicate under water for several years, other technologies are being developed based, e.g., on optical and radio-frequency electro-magnetic waves. Each technology has its own advantages and drawbacks: for example, acoustic signals achieve long communication ranges at order-of-kbit/s bit rate, whereas optical signals offer order-of-Mbit/s transmission rates but only over short transmitter--receiver distances. Such a technological diversity can be leveraged by multi-modal systems, which integrate different technologies and provide intelligence to decide which one should be used at any given time. In this paper, we address a fundamental part of this intelligence by proposing a novel routing protocol for networks of multi-modal nodes. The protocol makes distributed decisions about the flow in each link and over each technology at any given time, in order to advance a packet towards its destination. Our routing protocol prevents bottlenecks and allocates resources fairly to different nodes. We analyze the performance of our protocol via simulations and in a field experiment. The results show that our protocol successfully leverages all technologies to deliver data, even in the presence of imperfect topology information. To permit the reproduction of our results, we share our simulation code.
△ Less
Submitted 14 November, 2016;
originally announced November 2016.
-
Computationally Efficient Calculations of Target Performance of the Normalized Matched Filter Detector for Hydrocoustic Signals
Authors:
Roee Diamant
Abstract:
Detection of hydroacoustic transmissions is a key enabling technology in applications such as depth measurements, detection of objects, and undersea map**. To cope with the long channel delay spread and the low signal-to-noise ratio, hydroacoustic signals are constructed with a large time-bandwidth product, $N$. A promising detector for hydroacoustic signals is the normalized matched filter (NMF…
▽ More
Detection of hydroacoustic transmissions is a key enabling technology in applications such as depth measurements, detection of objects, and undersea map**. To cope with the long channel delay spread and the low signal-to-noise ratio, hydroacoustic signals are constructed with a large time-bandwidth product, $N$. A promising detector for hydroacoustic signals is the normalized matched filter (NMF). For the NMF, the detection threshold depends only on $N$, thereby obviating the need to estimate the characteristics of the sea ambient noise which are time-varying and hard to estimate. While previous works analyzed the characteristics of the normalized matched filter (NMF), for hydroacoustic signals with large $N$ values the expressions available are computationally complicated to evaluate. Specifically for hydroacoustic signals of large $N$ values, this paper presents approximations for the probability distribution of the NMF. These approximations are found extremely accurate in numerical simulations. We also outline a computationally efficient method to calculate the receiver operating characteristic (ROC) which is required to determine the detection threshold. Results from an experiment conducted in the Mediterranean sea at depth of 900~m agree with the analysis.
△ Less
Submitted 24 February, 2016;
originally announced April 2016.
