Training Sequence Research Articles

Generative Modeling of RNA Sequence Families with Restricted Boltzmann Machines.

In this chapter, we discuss the potential application of Restricted Boltzmann machines (RBM) to model sequence families of structured RNA molecules. RBMs are a simple two-layer machine learning model able to capture intricate sequence dependencies induced by secondary and tertiary structure, as well as mechanisms of structural flexibility, resulting in a model that can be successfully used for the design of allosteric RNA such as riboswitches. They have recently been experimentally validated as generative models for the SAM-I riboswitch aptamer domain sequence family. We introduce RBM mathematically and practically, providing self-contained code examples to download the necessary training sequence data, train the RBM, and sample novel sequences. We present in detail the implementation of algorithms necessary to use RBMs, focusing on applications in biological sequence modeling.

Enhanced frame synchronization and carrier recovery in coherent FSO communication: a pseudo-random and cyclic QPSK approach.

To alleviate the impact of atmospheric turbulence on receiver carrier recovery in free-space optical communication, this study introduces a novel frame synchronization and carrier recovery method. This method employs a blend of pseudorandom sequences and specific cyclic quadrature phase shift keying (QPSK) training sequences, facilitating frame synchronization and frequency offset (FO) estimation. The research designs a time-series metric curve to counteract the effect of side peaks, enhancing the accuracy of frequency offset estimation via QPSK spectrum features. Furthermore, the method incorporates a two-stage frequency offset estimation process, utilizing the inherent sign characteristics of the x and y polarization states. Applied to the quadrature amplitude modulation system under atmospheric turbulence conditions, the proposed algorithm demonstrates high precision under both strong and weak turbulence conditions. Furthermore, we confirmed the universality of the algorithm across multiple modulation formats.

MNSS: Neural Supersampling Framework for Real-Time Rendering on Mobile Devices.

Although neural supersampling has achieved great success in various applications for improving image quality, it is still difficult to apply it to a wide range of real-time rendering applications due to the high computational power demand. Most existing methods are computationally expensive and require high-performance hardware, preventing their use on platforms with limited hardware, such as smartphones. To this end, we propose a new supersampling framework for real-time rendering applications to reconstruct a high-quality image out of a low-resolution one, which is sufficiently lightweight to run on smartphones within a real-time budget. Our model takes as input the renderer-generated low resolution content and produces high resolution and anti-aliased results. To maximize sampling efficiency, we propose using an alternate sub-pixel sample pattern during the rasterization process. This allows us to create a relatively small reconstruction model while maintaining high image quality. By accumulating new samples into a high-resolution history buffer, an efficient history check and re-usage scheme is introduced to improve temporal stability. To our knowledge, this is the first research in pushing real-time neural supersampling on mobile devices. Due to the absence of training data, we present a new dataset containing 57 training and test sequences from three game scenes. Furthermore, based on the rendered motion vectors and a visual perception study, we introduce a new metric called inter-frame structural similarity (IF-SSIM) to quantitatively measure the temporal stability of rendered videos. Extensive evaluations demonstrate that our supersampling model outperforms existing or alternative solutions in both performance and temporal stability.

Predicting Alzheimer's Disease Progression Using a Versatile Sequence-Length-Adaptive Encoder-Decoder LSTM Architecture.

Detecting Alzheimer's disease (AD) accurately at an early stage is critical for planning and implementing disease-modifying treatments that can help prevent the progression to severe stages of the disease. In the existing literature, diagnostic test scores and clinical status have been provided for specific time points, and predicting the disease progression poses a significant challenge. However, few studies focus on longitudinal data to build deep-learning models for AD detection. These models are not stable to be relied upon in real medical settings due to a lack of adaptive training and testing. We aim to predict the individual's diagnostic status for the next six years in an adaptive manner where prediction performance improves with the number of patient visits. This study presents a Sequence-Length Adaptive Encoder-Decoder Long Short-Term Memory (SLA-ED LSTM) deep-learning model on longitudinal data obtained from the Alzheimer's Disease Neuroimaging Initiative archive. In the suggested approach, decoder LSTM dynamically adjusts to accommodate variations in training sequence length and inference length rather than being constrained to a fixed length. We evaluated the model performance for various sequence lengths and found that for inference length one, sequence length nine gives the highest average test accuracy and area under the receiver operating characteristic curves of 0.920 and 0.982, respectively. This insight suggests that data from nine visits effectively captures meaningful cognitive status changes and is adequate for accurate model training. We conducted a comparative analysis of the proposed model against state-of-the-art methods, revealing a significant improvement in disease progression prediction over the previous methods.

Exact Analysis of MIMO Channel Estimation Based on Superimposed Training

In this paper, channel estimation capabilities of a multiple-input multiple-output (MIMO) system using superimposed training sequences are investigated. A new expression for estimation-error variance is derived. It is shown that the training sequences must be balanced and must have specific correlation properties. The latter are required only in a specific zone. Sequences that satisfy these criteria exist and are referred to as zero-correlation zone (ZCZ) solutions. Consequently, by using balanced ZCZ sequences, harmful direct current (DC) offset can be removed. Owing to their zero-cross correlation, interference from other transmitting antennas may be eliminated. Furthermore, a closed-form expression of the estimation-error variance can be obtained due to their impulse-like autocorrelation. To increase the number of antennas in the MIMO system, a new construction of ZCZ sequence set is proposed, in which all sequences are balanced

How to Implement Authentic Scenario-Based Training

When used in the UK, scenarios are commonly positioned at the end of a linear training sequence where they are often used to test skills after the isolated practice of those skills has been completed. In contrast to the above view, authentic SBT is an approach based on concepts, tactics, and strategies, rather than skill. The focus is on introducing increasingly complex ‘whole’ scenarios (i.e., radio-call to event resolution) as training progresses. SBT has been shown to improve officer performance under pressure when high-fidelity (i.e., realistic) scenarios are used in this way in integrated de-escalation and use-of-force training. The challenges to implement authentic SBT are significant. When used in inappropriate ways, SBT will struggle to gain acceptance. However, when practice is premised on authentic SBT, it can be an important catalyst for changes to training.

Surgical training and research in mesh complications

Objective:To review and summarize the peer-reviewed literature for Surgical Training and Research in Mesh Complications. Methods:An international author team reviewed available published literature. Results:This paper reviews the current training situation and recognizes there are no fellowships globally which specifically train in mesh removal/ complications. The UK has recently launched a mesh complications training pathway pilot. New Zealand has recently implemented a credentialing framework for pelvic floor procedures including mesh removal and undertaken national credentialing of surgeons. A proposed training pathway is outlined including training sequence, the training setting, and a modular pathway of training.The authors acknowledge that there is very limited research and evidence for management of mesh complications. Several important research questions are suggested for future research. Future projects for collaborative research are proposed including registries, and development of improved patient education materials for pelvic floor conditions to improve shared decision making.

Enabling endogenous distributed acoustic sensing in a digital subcarrier coherent transmission system.

To monitor the health of the fiber network and its ambient environment in densely populated access/metro network areas, in this Letter, an endogenous distributed acoustic sensing (DAS) has been proposed and achieved in a coherent digital subcarrier multiplexing (DSCM) system. Rather than specially allocating a sensing probe in general integrated communication and sensing schemes, the fractional Fourier transformed (FrFT) training sequence (TS) designated for time/frequency synchronization in DSCM coherent communications has been repurposed for sensing. While achieving excellent synchronization performance of communication, the FrFT-based TS can also be concurrently utilized to perform distributed vibration sensing. Experimental results demonstrate that the FrFT-based timing/frequency synchronization sequence is repurposed to achieve a DAS sensitivity of 70 p ε/Hz at a spatial resolution of 5 m, along with 100-Gb/s 16 quadrature amplitude modulation (QAM) DSCM transmission, without a loss of spectral efficiency.

Intelligent forecasting temperature measurements of solar PV cells using modified recurrent neural network

For microgrids to operate optimally and minimize the effects of uncertainty, anticipating solar PV measurements is essential. For residential and commercial microgrids that use solar PV, the predicting of solar energy over a short period is crucial for managing grid-connected PVeffectively. Therefore, this work develops a Recurrent Neural Network (RNN) for forecasting temperature measurements as time series records, where a combination of long short-term memory (LSTM) architecture with RNN is used to process input measurements by updating the RNN state and winding over time degrees. Data from the entire prior time steps is stored in the RNN state. A dataset of temperature waveform measurements is used, which includes 2000 unnaturally produced signals of three channels with varying length. An LSTM neural network can be used to expect future values of a time series or sequence utilizing data from earlier time steps as input. Training of a regression LSTM neural network through the output of a sequence is performed, where the goals are the training sequence with records shifting one-time step, for training theLSTM neural architecture with time series forecasting. In other words, the weights of the LSTM neural structure learn to predict the following time step values of the input sequence at every time step. By considering the past forecasts as inputs, the closed-loop prediction forecasts the next time steps of sequences. The model makes the forecast without using the true data. The cross-entropy loss serves as the loss function. It is found that the mean RMSE overall test observations were about 0.5080 which promises to make better predictions from learning the temporal context of input sequences

RACER-m leverages structural features for sparse T cell specificity prediction.

Reliable prediction of T cell specificity against antigenic signatures is a formidable task, complicated by the immense diversity of T cell receptor and antigen sequence space and the resulting limited availability of training sets for inferential models. Recent modeling efforts have demonstrated the advantage of incorporating structural information to overcome the need for extensive training sequence data, yet disentangling the heterogeneous TCR-antigen interface to accurately predict MHC-allele-restricted TCR-peptide interactions has remained challenging. Here, we present RACER-m, a coarse-grained structural model leveraging key biophysical information from the diversity of publicly available TCR-antigen crystal structures. Explicit inclusion of structural content substantially reduces the required number of training examples and maintains reliable predictions of TCR-recognition specificity and sensitivity across diverse biological contexts. Our model capably identifies biophysically meaningful point-mutant peptides that affect binding affinity, distinguishing its ability in predicting TCR specificity of point-mutants from alternative sequence-based methods. Its application is broadly applicable to studies involving both closely related and structurally diverse TCR-peptide pairs.

In “Tone” with dogs: exploring canine musicality

This study investigates the musical perception skills of dogs through playback experiments. Dogs were trained to distinguish between two different target locations based on a sequence of four ascending or descending notes. A total of 16 dogs of different breeds, age, and sex, but all of them with at least basic training, were recruited for the study. Dogs received training from their respective owners in a suitable environment within their familiar home settings. The training sequence consisted of notes [Do-Mi-Sol#-Do (C7-E7-G7#-C8; Hz frequency: 2093, 2639, 3322, 4186)] digitally generated as pure sinusoidal tones. The training protocol comprised 3 sequential training levels, with each level consisting of 4 sessions with a minimum of 10 trials per session. In the test phase, the sequence was transposed to evaluate whether dogs used relative pitch when identifying the sequences. A correct response by the dog was recorded as 1, while an incorrect response, occurring when the dog chose the opposite zone of the bowl, was marked as 0. Statistical analyses were performed using a binomial test. Among 16 dogs, only two consistently performed above the chance level, demonstrating the ability to recognize relative pitch, even with transposed sequences. This study suggests that dogs may have the ability to attend to relative pitch, a critical aspect of human musicality.

FPGA implementation of joint synchronization algorithm based on short training sequence

Synchronization processing is a vital component of the Orthogonal Frequency Division Multiplexing (OFDM) receiver, playing a decisive role in determining the performance of the OFDM receiver. Based on the repetitive structure of the Short Training Sequence (STS) in the IEEE 802.11 physical layer protocol data unit, this paper proposes and implements a joint synchronization algorithm to address issues related to data packet detection and coarse frequency offset correction during the synchronization process. Through software simulation and Field-Programmable Gate Array (FPGA) board-level validation, it is concluded that the algorithm ensures performance while, to some extent, conserving FPGA resources.

ALGORITHM FOR IMPLEMENTATION OF MENTORSHIP SUPPORT FOR WAR VETERANS IN A TERRITORIAL COMMUNITY

In the article there have been studied issues and needs that war veterans and defenders of Ukraine have after demobilization from the service and return to civil life as well as flaws of the acting system of social protection for combatants and persons with disability caused by war. There have been presented the results of implementation of the social project «Mentors for veterans», which has been carried out by the non-governmental organization «Youth institute» in Uzyn municipal territorial community (Kyiv region) in 2023–2024. Under this project, there has been developed and tested the implementation mechanism for mentorship (assistance) institution for veterans at the level of the territorial community, that, at the same time, requires the implementation of three interdependent components, such as: informing local authorities and citizens of the community about the existing problem in transition of veterans to civil life and possible solutions; selection and training of mentors / assistants for veterans; implementation of the mentorship support for veterans on the level of the territorial community. In the article one can find a sequence order for development of a reference book, which includes reliable data about facilities and institutions that work with veterans, defenders of Ukraine and members of their families in the territorial community. This reference book will be useful for specialists in the social sphere during development of the system for information and mentorship support of military servants, who have been dismissed due to health conditions and / or family circumstances and undergo transition to civil life. In the article the sequence of involvement and training of mentors for veterans, contents of the training program for mentors have been explained. The algorithm of mentorship support for veterans, defenders of Ukraine and their family members in the territorial community that has been developed within the framework of the project covers the organization of the interaction between a mentor / an assistant and a veteran in the following order: establishment of contacts with experts in the social sphere, who work with veterans; the first meeting with the veteran (acquaintance, collection of information on needs / issues of veterans; keeping in touch / mentorship support of veterans); support in case of problematic situations; finalization of interaction with the veteran. The algorithm of mentorship support for veterans has been implemented in Uzyn municipal territorial community (Kyiv region) and that has allowed identifying the main conditions for efficient implementation of the mentorship (assistance) institution for veterans in the territorial community. Methodical and organizational forms developed and tested within the project frame will be useful for specialists in the social sphere during development of methodical recommendations and implementation of the assistance institution in the period of transition from military service to civil life and provision of the appropriate social services.

Physical layer security scheme for key concealment and distribution based on carrier scrambling

The purpose of this study is to present a physical layer security scheme for key concealment and distribution based on carrier scrambling. The three-dimensional (3D) Lorenz system is used to generate independent chaotic sequences that encrypt the information with bit, constellation and subcarrier. In order to realize the flexible distribution of the key and ensure its security, the key information is loaded into a specific subcarrier. While key subcarrier and the ciphertext subcarrier are scrambled simultaneously. The encrypted key position information is processed and transmitted in conjunction with the training sequence (TS) to facilitate demodulation by the legitimate receiver. The processed TS can accommodate up to 10 key position information, thereby demonstrating the scheme's exceptional scalability. Experimental results show that the proposed scheme can safely transmit 131.80 Gb/s Orthogonal frequency division multiplexing (OFDM) signals across 2 km 7-core fiber. Meanwhile, the scheme enables simultaneous flexible distribution and concealment of the key, thereby offering a promising solution for physical layer security.

Microscopic railway capacity assessment of heterogeneous traffic under real-life operational conditions

A key strategy for increasing railway capacity is utilizing infrastructure more efficiently. While much research has been completed on methods for assessing railway capacity, very little has focused on the details of capacity utilisation, such as assessing the various ways trains use capacity, the impacts of specific blocking time components, and how train dynamics (accelerating, cruising, braking, and dwelling) affect capacity.This paper presents a methodology for comparing planned occupancy to actual occupancy under real operations and applies it in a case study. The methodology is based on identifying a critical path which represents an extension of bottleneck concept presented in UIC leaflet 406. The methodology was applied in a case study to determine the specific blocking times and train dynamics which cause a blocking time gap for a sequence of trains, both a-priori and a-posteriori, after considering the operational variations. The analysis of real operations with variations in train trajectories shows that capacity occupation is mostly influenced by train sequence heterogeneity in the original schedule. The varying effects of operations have a smaller but relevant impact. The methods developed in this paper can be used to help assess railway capacity under real operations.

Autorotation Transfer of Training: Effects of Helicopter Dynamics

This paper analyzes the effects of the helicopter dynamics on pilots’ learning process and transfer of learned skills during autorotation training. A quasi-transfer-of-training experiment was performed with 10 experienced helicopter pilots in the SIMONA moving-base flight simulator at Delft University of Technology. Pilots had to control an in-house flight dynamics model setup to simulate two types of helicopter dynamics: (1) a “hard” dynamics characterized by a low autorotative flare index requiring high pilot control compensation and (2) a “easy” dynamics characterized by a high autorotative flare index with low pilot control compensation required. Two groups of pilots tested these types of dynamics in a different training sequence: hard-easy-hard (HEH group) and easy-hard-easy (EHE group). The main conclusion of this study proved that simulator training for autorotation can best start with pilots training in the most resource demanding condition. A more challenging helicopter’s dynamics will require higher pilot agility and more rapid responses to his/her perceptual changes. This will result in pilots developing more robust and adaptable flying skills. Indeed, a clear positive transfer of training effect was observed in the experiment presented in this paper in terms of acquired pilot skills in the HEH group, but not the EHE group. Positive transfer was especially observed in terms of reduced rate of descent at touchdown. The two groups differed in the control strategy applied, with the HEH group having developed a control technique mimicking more closely the one adopted in a real helicopter.

Universal Neyman–Pearson classification with a partially known hypothesis

Abstract We propose a universal classifier for binary Neyman–Pearson classification where the null distribution is known, while only a training sequence is available for the alternative distribution. The proposed classifier interpolates between Hoeffding’s classifier and the likelihood ratio test and attains the same error probability prefactor as the likelihood ratio test, i.e. the same prefactor as if both distributions were known. In addition, such as Hoeffding’s universal hypothesis test, the proposed classifier is shown to attain the optimal error exponent tradeoff attained by the likelihood ratio test whenever the ratio of training to observation samples exceeds a certain value. We propose a lower bound and an upper bound to the optimal training to observation ratio. In addition, we propose a sequential classifier that attains the optimal error exponent tradeoff.

Integrating virtual reality technology into beginning welder training sequences

Virtual reality (VR) technology is an advanced modern resource commonly integrated into various forms of training. VR training simulations are customizable in that quality-grading parameter settings, physical environment, and user capacity can be modified to personal or professional preference. In this study, VR technology training practices are utilized to enact meaningful learning. Meaningful learning is achieved by providing visual and audial cues within the virtual training environment, weld performance skill development, and adequate skill practice time over a four-week span. This method of practice will reflect a new training style where beginning welders receive personalized feedback from both the VRTEX 360 virtual reality welding simulator and welding instructor. This training method benefits learners by expediting and enhancing their skill acquisition, adjusting their performance according to the various feedback they receive, and thereby experiencing meaningful learning. Results indicate that with each round of VR welding training participants’ test weld scores continuously increased, as well as consistently scoring 80% and higher. This enhanced performance of beginning welders implies VR welding training can effectively aid in developing complex welding skills. We recommend that future research investigate the effectiveness of parameter cues and total cost-savings of integrated VR technology into welding training methods.

Unsupervised Training Sequence Design: Efficient and Generalizable Agent Training

To train generalizable Reinforcement Learning (RL) agents, researchers recently proposed the Unsupervised Environment Design (UED) framework, in which a teacher agent creates a very large number of training environments and a student agent trains on the experiences in these environments to be robust against unseen testing scenarios. For example, to train a student to master the “stepping over stumps” task, the teacher will create numerous training environments with varying stump heights and shapes. In this paper, we argue that UED neglects training efficiency and its need for very large number of environments (henceforth referred to as infinite horizon training) makes it less suitable to training robots and non-expert humans. In real-world applications where either creating new training scenarios is expensive or training efficiency is of critical importance, we want to maximize both the learning efficiency and learning outcome of the student. To achieve efficient finite horizon training, we propose a novel Markov Decision Process (MDP) formulation for the teacher agent, referred to as Unsupervised Training Sequence Design (UTSD). Specifically, we encode salient information from the student policy (e.g., behaviors and learning progress) into the teacher's state space, enabling the teacher to closely track the student's learning progress and consequently discover the optimal training sequences with finite lengths. Additionally, we explore the teacher's efficient adaptation to unseen students at test time by employing the context-based meta-learning approach, which leverages the teacher's past experiences with various students. Finally, we empirically demonstrate our teacher's capability to design efficient and effective training sequences for students with varying capabilities.

ESRL: Efficient Sampling-Based Reinforcement Learning for Sequence Generation

Applying Reinforcement Learning (RL) to sequence generation models enables the direct optimization of long-term rewards (e.g., BLEU and human feedback), but typically requires large-scale sampling over a space of action sequences. This is a computational challenge as presented by the practice of sequence generation problems, such as machine translation, where we often deal with a large action space (e.g., a vocabulary) and a long action sequence (e.g., a translation). In this work, we introduce two-stage sampling and dynamic sampling approaches to improve the sampling efficiency during training sequence generation models via RL. We experiment with our approaches on the traditional sequence generation tasks, including machine translation and abstractive summarization. Furthermore, we evaluate our approaches in RL from human feedback (RLHF) through training a large language model using the reward model. Experimental results show that the efficient sampling-based RL, referred to as ESRL, can outperform all baselines in terms of both training efficiency and memory consumption. Notably, ESRL yields consistent performance gains over the strong REINFORCE, minimum risk training, and proximal policy optimization methods. The code is available at https://github.com/wangclnlp/DeepSpeed-Chat-Extension/examples/esrl.