Search

A common challenge towards the adaptability of Large Language Models (LLMs) is their ability to learn new languages over time without hampering the model's performance on languages in which the model is already proficient (usually English). Continual fine-tuning (CFT) is the process of sequentially fine-tuning an LLM to enable the model to adapt to downstream tasks with varying data distributions and time shifts. This paper focuses on the language adaptability of LLMs through CFT. We study a two-phase CFT process in which an English-only end-to-end fine-tuned LLM from Phase 1 (predominantly Task Ability) is sequentially fine-tuned on a multilingual dataset -- comprising task data in new languages -- in Phase 2 (predominantly Language Ability). We observe that the ``similarity'' of Phase 2 tasks with Phase 1 determines the LLM's adaptability. For similar phase-wise datasets, the LLM after Phase 2 does not show deterioration in task ability. In contrast, when the phase-wise datasets are not similar, the LLM's task ability deteriorates. We test our hypothesis on the open-source \mis\ and \llm\ models with multiple phase-wise dataset pairs. To address the deterioration, we analyze tailored variants of two CFT methods: layer freezing and generative replay. Our findings demonstrate their effectiveness in enhancing the language ability of LLMs while preserving task performance, in comparison to relevant baselines., Comment: 19 pages, 6 tables, 4 figures

Low latency models are critical for real-time speech enhancement applications, such as hearing aids and hearables. However, the sub-millisecond latency space for resource-constrained hearables remains underexplored. We demonstrate speech enhancement using a computationally efficient minimum-phase FIR filter, enabling sample-by-sample processing to achieve mean algorithmic latency of 0.32 ms to 1.25 ms. With a single microphone, we observe a mean SI-SDRi of 4.1 dB. The approach shows generalization with a DNSMOS increase of 0.2 on unseen audio recordings. We use a lightweight LSTM-based model of 644k parameters to generate FIR taps. We benchmark that our system can run on low-power DSP with 388 MIPS and mean end-to-end latency of 3.35 ms. We provide a comparison with baseline low-latency spectral masking techniques. We hope this work will enable a better understanding of latency and can be used to improve the comfort and usability of hearables.

We demonstrate spectral shaping of entangled photons with a six-channel microring-resonator-based silicon photonic pulse shaper. Through precise calibration of thermal phase shifters in a microresonator-based pulse shaper, we demonstrate line-by-line phase control on a 3~GHz grid for two frequency-bin-entangled qudits, corresponding to Hilbert spaces of up to $6\times 6$ ($3\times 3$) dimensions for shared (independent) signal-idler filters. The pulse shaper's fine spectral resolution enables control of nanosecond-scale temporal features, which are observed by direct coincidence detection of biphoton correlation functions that show excellent agreement with theory. This work marks, to our knowledge, the first demonstration of biphoton pulse shaping using an integrated spectral shaper and holds significant promise for applications in quantum information processing.

The identification of persistent radio sources (PRSs) coincident with two repeating fast radio bursts (FRBs) supports FRB theories requiring a compact central engine. However, deep non-detections in other cases highlight the diversity of repeating FRBs and their local environments. Here, we perform a systematic search for radio sources towards 37 CHIME/FRB repeaters using their arcminute localizations and a combination of archival surveys and targeted observations. Through multi-wavelength analysis of individual radio sources, we identify two (20181030A-S1 and 20190417A-S1) for which we disfavor an origin of either star formation or an active galactic nucleus in their host galaxies and thus consider them candidate PRSs. We do not find any associated PRSs for the majority of the repeating FRBs in our sample. For 8 FRB fields with Very Large Array imaging, we provide deep limits on the presence of PRSs that are 2--4 orders of magnitude fainter than the PRS associated with FRB\,20121102A. Using Very Large Array Sky Survey imaging of all 37 fields, we constrain the rate of luminous ($\gtrsim$10$^{40}$ erg s$^{-1}$) PRSs associated with repeating FRBs to be low. Within the context of FRB-PRS models, we find that 20181030A-S1 and 20190417A-S1 can be reasonably explained within the context of magnetar, hypernebulae, gamma-ray burst afterglow, or supernova ejecta models -- although we note that both sources follow the radio luminosity versus rotation measure relationship predicted in the nebula model framework. Future observations will be required to both further characterize and confirm the association of these PRS candidates with the FRBs., Comment: 37 pages, 12 figures

Collaborating in a group, whether face-to-face or virtually, involves continuously expressing emotions and interpreting those of other group members. Therefore, understanding group affect is essential to comprehending how groups interact and succeed in collaborative efforts. In this study, we move beyond individual-level affect and investigate group-level affect -- a collective phenomenon that reflects the shared mood or emotions among group members at a particular moment. As the first in literature, we gather annotations for group-level affective expressions using a fine-grained temporal approach (15 second windows) that also captures the inherent dynamics of the collective construct. To this end, we use trained annotators and an annotation procedure specifically tuned to capture the entire scope of the group interaction. In addition, we model group affect dynamics over time. One way to study the ebb and flow of group affect in group interactions is to model the underlying convergence (driven by emotional contagion) and divergence (resulting from emotional reactivity) of affective expressions amongst group members. To capture these interpersonal dynamics, we extract synchrony based features from both audio and visual social signal cues. An analysis of these features reveals that interacting groups tend to diverge in terms of their social signals along neutral levels of group affect, and converge along extreme levels of affect expression. We further present results on the predictive modeling of dynamic group affect which underscores the importance of using synchrony-based features in the modeling process, as well as the multimodal nature of group affect. We anticipate that the presented models will serve as the baselines of future research on the automatic recognition of dynamic group affect., Comment: This work has been submitted to the IEEE for possible publication

The uncertainties in photometric redshifts and stellar masses from imaging surveys affect galaxy sample selection, their abundance measurements, as well as the measured weak lensing signals. We develop a framework to assess the systematic effects arising from the use of redshifts and stellar masses derived from photometric data, and explore their impact on the inferred galaxy-dark matter connection. We use galaxy catalogues from the UniverseMachine (UM) galaxy formation model to create Pz-mock galaxy samples that approximately follow the redshift errors in the Subaru HSC survey. We focus on galaxy stellar-mass thresholds ranging from $\log\left[M_*/(h^{-2}M_\odot)\right]$ from $8.6$ to $11.2$ in steps of 0.2 dex within two redshift bins $0.30-0.55$ and $0.55-0.80$. A comparison of the Pz-mock samples to true galaxy samples in UM shows a relatively mild sample contamination for thresholds with $\log\left[M_{*,\rm limit}/(h^{-2}M_\odot)\right]<10.6$, while an increasing contamination towards the more massive end. We show how such contamination affects the measured abundance and the lensing signal. A joint HOD modelling of the observables from the Pz-mock compared to the truth in the UM informs the systematic biases on the average halo masses of central galaxies in the HSC survey. Even with a reasonably conservative choice of photo-$z$ errors in Pz-mock, we show that the inferred halo masses deduced from the HSC galaxies for low-mass thresholds will have a systematic bias smaller than 0.05 dex. Beyond $\log\left[M_{*,\rm limit}/(h^{-2}M_\odot)\right]=10.6$, the inferred halo masses show an increasing systematic bias with stellar mass, reaching values of order $0.2$ dex, larger than the statistical error., Comment: 20 pages, 20 figures, Submitted to MNRAS Journal

This study introduces a novel framework to apply Artifact Subspace Reconstruction (ASR) algorithm on single-channel Electroencephalogram (EEG) data. ASR, renowned for its automated capability to effectively eliminate various artifacts like eye-blinks and eye movements from EEG signals. Importantly it has been implemented on android smartphones, but relied on multiple channels for principal component subspace calculations. To overcome this limitation, we incorporate the established dynamical embedding approach into the algorithm, naming it Embedded-ASR (E-ASR). In our proposed method, an embedded matrix is first constructed from a single-channel EEG data using series of delay vectors. ASR is then applied to this embedded matrix, and the resulting cleaned single-channel EEG is reconstructed by removing the time lag and concatenating the rows of the embedded matrix. Data was collected from four subjects in resting states with eyes open from pre-frontal (Fp1 and Fp2) electrodes using CameraEEG app. To assess the effectiveness of the E-ASR algorithm on an EEG dataset with artifacts, we employed performance metrics such as relative root mean square error (RRMSE), correlation coefficient (CC), average power ratio as well as estimated the number of eye-blinks with and without the E-ASR approach. E-ASR was able to reduce artifacts from the semi-simulated EEG data, with an RRMSE of 45.45% and a CC of 0.91. For real EEG data, the counted eye-blinks were manually cross-checked with ground truth obtained from CameraEEG video data across all subjects for individual Fp1 and Fp2 electrodes. In conclusion, our study suggests E-ASR framework can remove artifacts from single channel EEG data. This promising algorithm might have potential for smartphone-based natural environment EEG applications, where minimal number of electrodes is a critical factor.

For multidimensional Euclidean type spaces, we study convex choice: from any choice set, the set of types that make the same choice is convex. We establish that, in a suitable sense, this property characterizes the sufficiency of local incentive constraints. Convex choice is also of interest more broadly. We tie convex choice to a notion of directional single-crossing differences (DSCD). For an expected-utility agent choosing among lotteries, DSCD implies that preferences are either one-dimensional or must take the affine form that has been tractable in multidimensional mechanism design.

Do elections efficiently aggregate politicians' policy-relevant private information? This paper argues that politicians' office motivation is an obstacle. In a two-candidate Hotelling-Downs model in which each candidate has socially-valuable policy information, we establish that equilibrium welfare is at best what can be obtained by disregarding one politician's information. We also find that for canonical information structures, politicians have an incentive to ``anti-pander'', i.e., to overreact to their information. Some degree of pandering -- underreacting to information -- would be socially beneficial.

Social media platforms, particularly Telegram, play a pivotal role in shaping public perceptions and opinions on global and national issues. Unlike traditional news media, Telegram allows for the proliferation of user-generated content with minimal oversight, making it a significant venue for the spread of controversial and misinformative content. During the COVID-19 pandemic, Telegram's popularity surged in Singapore, a country with one of the highest rates of social media use globally. We leverage Singapore-based Telegram data to analyze information flows within groups focused on COVID-19 and climate change. Using k-means clustering, we identified distinct user archetypes, including Skeptic, Engaged Advocate, Observer, and Analyst, each contributing uniquely to the discourse. We developed a model to classify users into these clusters (Precision: Climate change: 0.99; COVID-19: 0.95). By identifying these user archetypes and examining their contributions to information dissemination, we sought to uncover patterns to inform effective strategies for combating misinformation and enhancing public discourse on pressing global issues., Comment: Incomplete data and modification in data analysis

The aim of this study is to teach an algorithm how to recognize different types of music. Users will submit songs for analysis. Since the algorithm hasn't heard these songs before, it needs to figure out what makes each song unique. It does this by breaking down the songs into different parts and studying things like rhythm, melody, and tone via supervised learning because the program learns from examples that are already labelled. One important thing to consider when classifying music is its genre, which can be quite complex. To ensure accuracy, we use five different algorithms, each working independently, to analyze the songs. This helps us get a more complete understanding of each song's characteristics. Therefore, our goal is to correctly identify the genre of each submitted song. Once the analysis is done, the results are presented using a graphing tool, making it easy for users to understand and provide feedback.

The inner workings of neural networks can be better understood if we can fully decipher the information encoded in neural activations. In this paper, we argue that this information is embodied by the subset of inputs that give rise to similar activations. Computing such subsets is nontrivial as the input space is exponentially large. We propose InversionView, which allows us to practically inspect this subset by sampling from a trained decoder model conditioned on activations. This helps uncover the information content of activation vectors, and facilitates understanding of the algorithms implemented by transformer models. We present four case studies where we investigate models ranging from small transformers to GPT-2. In these studies, we demonstrate the characteristics of our method, show the distinctive advantages it offers, and provide causally verified circuits., Comment: ICML 2024 Mechanistic Interpretability Workshop oral

Enterprises grapple with the significant challenge of managing proprietary unstructured data, hindering efficient information retrieval. This has led to the emergence of AI-driven information retrieval solutions, designed to adeptly extract relevant insights to address employee inquiries. These solutions often leverage pre-trained embedding models and generative models as foundational components. While pre-trained embeddings may exhibit proximity or disparity based on their original training objectives, they might not fully align with the unique characteristics of enterprise-specific data, leading to suboptimal alignment with the retrieval goals of enterprise environments. In this paper, we propose a comprehensive methodology for contextualizing pre-trained embedding models to enterprise environments, covering the entire process from data preparation to model fine-tuning and evaluation. By adapting the embeddings to better suit the retrieval tasks prevalent in enterprises, we aim to enhance the performance of information retrieval solutions. We discuss the process of fine-tuning, its effect on retrieval accuracy, and the potential benefits for enterprise information management. Our findings demonstrate the efficacy of fine-tuned embedding models in improving the precision and relevance of search results in enterprise settings.

We investigate the information-theoretic conditions to achieve the complete reconstruction of a diploid genome. We also analyze the standard greedy and de-Bruijn graph-based algorithms and compare the coverage depth and read length requirements with the information-theoretic lower bound. Our results show that the gap between the two is considerable because both algorithms require the double repeats in the genome to be bridged., Comment: Accepted at ISIT'24

An m-uniform quantum state on n qubits is an entangled state in which every m-qubit subsystem is maximally mixed. Such a state spans a pure [[n,0,m+1]] quantum error correcting code (QECC). Starting with an m-uniform state realized as the graph state associated with an m-regular graph, and a classical [n,k,d \ge m+1] binary linear code with certain additional properties, we construct codeword stabilized (CWS) codes that are pure [[n,k,m+1]] QECCs. We propose measurement-based protocols for encoding into code states and recovery of logical qubits from code states. Our proposed protocols support sequential encoding and partial recovery of logical qubits, which can be useful for quantum memory applications., Comment: A shorter version of this manuscript is available in the Proceedings of the 2024 International Symposium on Information Theory (ISIT 2024)

A weakly constrained code is a collection of finite-length strings over a finite alphabet in which certain substrings or patterns occur according to some prescribed frequencies. Buzaglo and Siegel (ITW 2017) gave a construction of weakly constrained codes based on row-by-row coding, that achieved the capacity of the weak constraint. In this paper, we propose a method to make this row-by-row coding scheme resilient to errors., Comment: 9 pages, a shorter version is submitted at the International Symposium on Information Theory (ISIT) 2024

Transformers trained on natural language data have been shown to learn its hierarchical structure and generalize to sentences with unseen syntactic structures without explicitly encoding any structural bias. In this work, we investigate sources of inductive bias in transformer models and their training that could cause such generalization behavior to emerge. We extensively experiment with transformer models trained on multiple synthetic datasets and with different training objectives and show that while other objectives e.g. sequence-to-sequence modeling, prefix language modeling, often failed to lead to hierarchical generalization, models trained with the language modeling objective consistently learned to generalize hierarchically. We then conduct pruning experiments to study how transformers trained with the language modeling objective encode hierarchical structure. When pruned, we find joint existence of subnetworks within the model with different generalization behaviors (subnetworks corresponding to hierarchical structure and linear order). Finally, we take a Bayesian perspective to further uncover transformers' preference for hierarchical generalization: We establish a correlation between whether transformers generalize hierarchically on a dataset and whether the simplest explanation of that dataset is provided by a hierarchical grammar compared to regular grammars exhibiting linear generalization., Comment: Code now available: https://github.com/kabirahuja2431/transformers-hg

Rowlinson et al. 2023 recently claimed the detection of a coherent radio flash 76.6 minutes after a short gamma-ray burst. They proposed that the radio emission may be associated with a long-lived neutron star engine. We show through theoretical and observational arguments that the coherent radio emission, if real and indeed associated with GRB 201006A and at the estimated redshift, is unlikely to be due to the collapse of the neutron star, ruling out a blitzar-like mechanism. Instead, we show if a long-lived engine was created, it must have been stable with the radio emission likely linked to the intrinsic magnetar activity. However, we find that the optical upper limits require fine-tuning to be consistent with a magnetar-driven kilonova: we show that neutron-star engines that do satisfy the optical constraints would have produced a bright kilonova afterglow that should already be observable by the VLA or MeerKAT (for ambient densities typical for short GRBs). Given the optical limits and the current lack of a kilonova afterglow, we instead posit that no neutron star survived the merger, and the coherent radio emission was produced far from a black hole central engine via mechanisms such as synchrotron maser or magnetic reconnection in the jet -- a scenario consistent with all observations. We encourage future radio follow-up to probe the engine of this exciting event and continued prompt radio follow-up of short GRBs., Comment: Submitted. 10 pages, 4 figures

For Music Information Retrieval downstream tasks, the most common audio representation is time-frequency-based, such as Mel spectrograms. In order to identify musical genres, this study explores the possibilities of a new form of audio representation one of the most usual MIR downstream tasks. Therefore, to discretely encoding music using deep vector quantization; a novel audio representation was created for the innovative generative music model i.e. Jukebox. The effectiveness of Jukebox's audio representation is compared to Mel spectrograms using a dataset that is almost equivalent to State-of-the-Art (SOTA) and an almost same transformer design. The results of this study imply that, at least when the transformers are pretrained using a very modest dataset of 20k tracks, Jukebox's audio representation is not superior to Mel spectrograms. This could be explained by the fact that Jukebox's audio representation does not sufficiently take into account the peculiarities of human hearing perception. On the other hand, Mel spectrograms are specifically created with the human auditory sense in mind.

This study examines the effects of macroeconomic policies on financial markets using a novel approach that combines Machine Learning (ML) techniques and causal inference. It focuses on the effect of interest rate changes made by the US Federal Reserve System (FRS) on the returns of fixed income and equity funds between January 1986 and December 2021. The analysis makes a distinction between actively and passively managed funds, hypothesizing that the latter are less susceptible to changes in interest rates. The study contrasts gradient boosting and linear regression models using the Double Machine Learning (DML) framework, which supports a variety of statistical learning techniques. Results indicate that gradient boosting is a useful tool for predicting fund returns; for example, a 1% increase in interest rates causes an actively managed fund's return to decrease by -11.97%. This understanding of the relationship between interest rates and fund performance provides opportunities for additional research and insightful, data-driven advice for fund managers and investors

Video games have played a crucial role in entertainment since their development in the 1970s, becoming even more prominent during the lockdown period when people were looking for ways to entertain them. However, at that time, players were unaware of the significant impact that playtime could have on their feelings. This has made it challenging for designers and developers to create new games since they have to control the emotional impact that these games will take on players. Thus, the purpose of this study is to look at how a player's emotions are affected by the duration of the game. In order to achieve this goal, a framework for emotion detection is created. According to the experiment's results, the volunteers' general ability to express emotions increased from 20 to 60 minutes. In comparison to shorter gameplay sessions, the experiment found that extended gameplay sessions did significantly affect the player's emotions. According to the results, it was recommended that in order to lessen the potential emotional impact that playing computer and video games may have in the future, game producers should think about creating shorter, entertaining games.

Boger fluids are viscoelastic liquids having constant viscosity for a broad range of shear rates. They are commonly used to separate the effects of liquid elasticity from viscosity in any experiment. We present an experimental study on the shock-induced aerobreakup of a Boger fluid droplet in the Shear-induced entrainment (SIE) and catastrophic breakup regime (Weber number ranging from ~ 800 to 5000). The results are compared with the aerobreakup of a Newtonian droplet having similar viscosity, and with shear-thinning droplets. The study aims to identify the role of liquid elasticity without the added complexity of simultaneous shear-thinning behavior. It is observed that at the early stages of droplet breakup, liquid elasticity plays an insignificant role, and all the fluids show similar behavior. However, during the late stages, the impact of liquid elasticity becomes dominant, which results in a markedly different morphology of the fragmenting liquid mass compared to a Newtonian droplet.

This paper develops an algorithmic approach for obtaining estimates of the weight enumerators of Reed-Muller (RM) codes. Our algorithm is based on a technique for estimating the partition functions of spin systems, which in turn employs a sampler that produces codewords according to a suitably defined Gibbs distribution. We apply our method to moderate-blocklength RM codes and derive approximate values of their weight enumerators. We observe that the rates of the weight enumerator estimates returned by our method are close to the true rates when these rates are either known or computable by brute-force search; in other cases, our computations provide provably robust estimates. As a byproduct, our sampling algorithm also allows us to obtain estimates of the weight spectra of RM codes. We illustrate our methods by providing estimates of the hitherto unknown weight enumerators of the RM$(11,5)$ code and the exact weight spectra of the RM$(10,3)$ and RM$(10,4)$ codes., Comment: 8 pages, 1 figure, 4 tables; submitted to the IEEE for possible publication. arXiv admin note: substantial text overlap with arXiv:2309.08907

Optical pulse shaping stands as a formidable technique in ultrafast optics, radio-frequency photonics, and quantum communications. While existing systems rely on bulk optics or integrated platforms with planar waveguide sections for spatial dispersion, they face limitations in achieving finer (few- or sub-GHz) spectrum control. These methods either demand considerable space or suffer from pronounced phase errors and optical losses when assembled to achieve fine resolution. Addressing these challenges, we present a foundry-fabricated six-channel silicon photonic shaper using microresonator filter banks with inline phase control and high spectral resolution. Leveraging existing comb-based spectroscopic techniques, we devise a novel system to mitigate thermal crosstalk and enable the versatile use of our on-chip shaper. Our results demonstrate the shaper's ability to phase-compensate six comb lines at tunable channel spacings of 3, 4, and 5 GHz. Specifically, at a 3 GHz channel spacing, we showcase the generation of arbitrary waveforms in the time domain. This scalable design and control scheme holds promise in meeting future demands for high-precision spectral shaping capabilities., Comment: 21 pages, 12 figures

Audio fingerprinting, exemplified by pioneers like Shazam, has transformed digital audio recognition. However, existing systems struggle with accuracy in challenging conditions, limiting broad applicability. This research proposes an AI and ML integrated audio fingerprinting algorithm to enhance accuracy. Built on the Dejavu Project's foundations, the study emphasizes real-world scenario simulations with diverse background noises and distortions. Signal processing, central to Dejavu's model, includes the Fast Fourier Transform, spectrograms, and peak extraction. The "constellation" concept and fingerprint hashing enable unique song identification. Performance evaluation attests to 100% accuracy within a 5-second audio input, with a system showcasing predictable matching speed for efficiency. Storage analysis highlights the critical space-speed trade-off for practical implementation. This research advances audio fingerprinting's adaptability, addressing challenges in varied environments and applications.

Clinicians often investigate the body mass index (BMI) trajectories of children to assess their growth with respect to their peers, as well as to anticipate future growth and disease risk. While retrospective modelling of BMI trajectories has been an active area of research, prospective prediction of continuous BMI trajectories from historical growth data has not been well investigated. Using weight and height measurements from birth to age 10 years from a longitudinal mother-offspring cohort, we leveraged a multi-task Gaussian processes model, called MagmaClust, to derive probabilistic predictions for BMI trajectories over various forecasting periods. Experiments were conducted to evaluate the accuracy, sensitivity to missing values, and number of clusters. The results were compared with cubic B-spline regression and a parametric Jenss-Bayley mixed effects model. A downstream tool computing individual overweight probabilities was also proposed and evaluated. In all experiments, MagmaClust outperformed conventional models in prediction accuracy while correctly calibrating uncertainty regardless of the missing data amount (up to 90\% missing) or the forecasting period (from 2 to 8 years in the future). Moreover, the overweight probabilities computed from MagmaClust's uncertainty quantification exhibited high specificity ($0.94$ to $0.96$) and accuracy ($0.86$ to $0.94$) in predicting the 10-year overweight status even from age 2 years. MagmaClust provides a probabilistic non-parametric framework to prospectively predict BMI trajectories, which is robust to missing values and outperforms conventional BMI trajectory modelling approaches. It also clusters individuals to identify typical BMI patterns (early peak, adiposity rebounds) during childhood. Overall, we demonstrated its potential to anticipate BMI evolution throughout childhood, allowing clinicians to implement prevention strategies., Comment: 17 pages, 9 figures, 5 tables

End-to-end multi-task dialogue systems are usually designed with separate modules for the dialogue pipeline. Among these, the policy module is essential for deciding what to do in response to user input. This policy is trained by reinforcement learning algorithms by taking advantage of an environment in which an agent receives feedback in the form of a reward signal. The current dialogue systems, however, only provide meagre and simplistic rewards. Investigating intrinsic motivation reinforcement learning algorithms is the goal of this study. Through this, the agent can quickly accelerate training and improve its capacity to judge the quality of its actions by teaching it an internal incentive system. In particular, we adapt techniques for random network distillation and curiosity-driven reinforcement learning to measure the frequency of state visits and encourage exploration by using semantic similarity between utterances. Experimental results on MultiWOZ, a heterogeneous dataset, show that intrinsic motivation-based debate systems outperform policies that depend on extrinsic incentives. By adopting random network distillation, for example, which is trained using semantic similarity between user-system dialogues, an astounding average success rate of 73% is achieved. This is a significant improvement over the baseline Proximal Policy Optimization (PPO), which has an average success rate of 60%. In addition, performance indicators such as booking rates and completion rates show a 10% rise over the baseline. Furthermore, these intrinsic incentive models help improve the system's policy's resilience in an increasing amount of domains. This implies that they could be useful in scaling up to settings that cover a wider range of domains., Comment: 6 pages, 1 figure, 18th IEEE International Conference on Semantic Computing

We have a new paradigm to design NMR pulses. Pulses, we call feedback pulses. We want broadband inversion and excitation. We have many offsets, start evolving them all starting from the north pole. Monitor them on the Bloch sphere, see which offset is worst (most away from south pole). Change the rf-phase to the offset ($\pi/2$ ahead of offset transverse magnetization phase) and irradiate at that offset frequency and evolve for some time and monitor and repeat, looking for worst offset. When we are on resonance to a offset, we are doing well, inverting it and when we are off resonant, we don't hurt much (even if hurt little, we will come back to the offset in good time). By the process of monitoring, and setting phase we eventually push everything to the south pole and bingo, we have an inversion pulse. Feedback is done in simulation, but what results in end is a broadband inversion pulse. For broadband excitation, start with all offsets (symmetric around origin) on y axis. By feedback push them to the south pole. When we run the resulting sequence backward with phases, $\pi$ incremented, we will get an excitation pulse. For band-selective excitation pulse put offsets in pass band on the $y$ axis and in the stop band on the south pole. Use feedback to push everything to the south pole. Again, run backwards with $\pi$ incremented phases, to get band selective excitation. Suddenly, we have it all, simple and easy. The paper, introduces the feedback pulse algorithm, simulations and experiments., Comment: 16 pgs, 5 figures

Bees undergo a self-organised process known as shimmering, where they form emergent patterns when they interact with each other on the nest surface as a defence mechanism in response to predator attacks. Many experimental studies have empirically investigated how the transfer of information to neighbouring bees propagates in various shimmering processes by measuring shimmering wave strength. However, there is no analytical modelling of the collective defence mechanism in nature. Here we introduce the first analytical tri-state Inactive-Active-Relapse (IAR) model to formulate the intrinsic process of bee shimmering. The major shimmering behaviour is shown to emerge under theoretical conditions which is demonstrated numerically and visually by simulating 1,000,000 bee agents, while the number of agents is scalable. Furthermore, we elaborate on these mathematical results to construct a wave strength function to demonstrate the accuracy of shimmering dynamics. The constructed wave strength function can be adapted to peak between 50-150ms which supports the experimental studies. Our results provide a foundation for further theoretical understanding of bee shimmering wave dynamics and could serve as inspiration for modelling other self-organised phenomena across scientific applications., Comment: 20 pages, 7 figures