Inference Error Research Articles

Tiny and error-tolerant ConvLSTM for event-based vision sensor with optimized event representation and ReRAM computation-in-memory

This paper proposes a tiny and error-tolerant convolutional long short-term memory network (ConvLSTM) for event-based vision sensor data processing with computation-in-memory (CiM) that utilizes resistive random-access memory (ReRAM). To handle the event data by ConvLSTM, the type and parameter of event representation are optimized to pre-process event data. To realize tiny and error-tolerant ConvLSTM CiM, reductions of the bit precision and error tolerance in each ConvLSTM layer and weight are considered. On the other hand, reducing the bit precision degrades inference accuracy and error tolerance. ReRAM has non-idealities such as conductance variation and wearing out by Set/Reset cycles that induces bit inversion. Thus, the bit error rate at each Set/Reset cycle is investigated, and the bit precision of the weight is optimized to accept a bit error rate of 0.1% in simulation. The proposed tiny ConvLSTM can reduce the size of memory by 91% compared with non-optimized ConvLSTM for which weight is represented in 32-bit precision.

A Crowd-Aided Vehicular Hybrid Sensing Framework for Intelligent Transportation Systems

The development of Internet of Things (IoT) techniques enables the paradigm shift in traffic data collecting. In traditional practices of transportation system's constructions, traffic-related information is collected based on dedicated sensor networks, which are not only coverage-limited but also cost-consuming. With the enrichment of the concepts concerning “social sensors” and “social transportation”, Sparse mobile crowdsensing (MCS) emerges as a promising sensing paradigm to collect data from only a few subareas by recruiting vehicles or mobile users with portable devices and to infer the data in unsensed subareas with acceptable errors at a low-cost manner. However, in real-world sensing campaigns, the Sparse MCS systems often fail to collect data from any subareas of interest since the assumption about sufficient participants is not always realistic. To be specific, the recruitment of participants is often limited by interest deficiency, privacy awareness, and distribution biases. To handle this problem, we introduce the dedicated sensing vehicles (DSVs) e.g., drones or driverless vehicles into traditional Sparse MCS to improve subarea coverage and inference performance. To achieve effective collaboration among DSVs and mobile users, we first design a crowd-aided vehicular hybrid sensing framework, which defines the order of task assignment for DSVs and mobile users as well as the budget allocation. In terms of DSVs route planning, we propose a three-step strategy, including optimal route searching, fused route selection, and final route determination. Moreover, mobile users are selected based on a proposed novel selection strategy. Experimental findings on two real-world datasets validate the effectiveness (with less inference error) of the hybrid sensing framework and the proposed strategies, in comparison with the user-only/DSV-only framework and five baselines. Results reveal important implications of applying the hybrid sensing paradigm in intelligent transportation systems to enhance data collection.

A Method for Detecting Outliers from the Gamma Distribution

Outliers often occur during data collection, which could impact the result seriously and lead to a large inference error; therefore, it is important to detect outliers before data analysis. Gamma distribution is a popular distribution in statistics; this paper proposes a method for detecting multiple upper outliers from gamma (m,θ). For computing the critical value of the test statistic in our method, we derive the density function for the case of a single outlier and design two algorithms based on the Monte Carlo and the kernel density estimation for the case of multiple upper outliers. A simulation study shows that the test statistic proposed in this paper outperforms some common test statistics. Finally, we propose an improved testing method to reduce the impact of the swamping effect, which is demonstrated by real data analyses.

Qubit readout and quantum sensing with pulses of quantum radiation

Different hypotheses about a quantum system such as the logical state of a qubit or the value of physical interaction parameters can be investigated by the interaction with a probe field. Such fields may be prepared in particularly sensitive quantum states and we demonstrate here the use of quantum trajectories to model the stochastic measurement record and conditional evolution of the state of the quantum system subject to its interaction with a traveling pulse of radiation. Our analysis applies to different measurement strategies and to arbitrary input quantum states of the probe field pulse and it thus permits direct comparison of their metrological advantages. A theoretical lower limit to the mean discrimination error can be calculated in a deterministic manner and we verify that it lies below the average inference error in all our examples.

Analysis of subjective thermal comfort data: A statistical point of view

Thermal comfort research aims to determine the relationship between the thermal environment and the human sense of warmth. This is usually achieved by measuring the subjective human thermal response to different thermal environments. However, it is common practice to use simple linear regression to analyse data collected using ordinal scales. This practice may lead to severe errors in inference. This study first set the methodological foundations to analyse subjective thermal comfort data from a statistical perspective. Subsequently, we show the practical consequences of fallacious assumptions by utilising a Bayesian approach and show, through an illustrative example, that a linear regression model applied to ordinal data suggests results different from those obtained using ordinal regression. Specifically, linear regression found no difference in means and effect size between genders, while the ordinal regression model led to the opposite conclusion. In addition, the linear regression model distorts the estimated regression coefficient for air temperature compared to the ordinal model. Finally, the ordinal model shows that the distance between adjacent response categories of the ASHRAE 7-point thermal sensation scale is not equidistant. Given the abovementioned issues, we advocate utilising ordinal models instead of metric models to analyse ordinal data.

Pay Attention via Quantization: Enhancing Explainability of Neural Networks via Quantized Activation

Modern deep learning algorithms comprise highly complex artificial neural networks, making it extremely difficult for humans to track their inference processes. As the social implementation of deep learning progresses, the human and economic losses caused by inference errors are becoming increasingly problematic, making it necessary to develop methods to explain the basis for the decisions of deep learning algorithms. Although an attention mechanism-based method to visualize the regions that contribute to steering angle prediction in an automated driving task has been proposed, its explanatory capability is low. In this paper, we focus on the fact that the importance of each bit in the activation value of a network is biased (i.e., the sign and exponent bits are weighted more heavily than the mantissa bits), which has been overlooked in previous studies. Specifically, this paper quantizes network activations, encouraging important information to be aggregated to the sign bit. Further, we introduce an attention mechanism restricted to the sign bit to improve the explanatory power. Our numerical experiment using the Udacity dataset revealed that the proposed method achieves a 1.14× higher area under curve (AUC) in terms of the deletion metric.

Machine Learning and Prediction Errors in Causal Inference

Machine Learning and Prediction Errors in Causal Inference

Methods for reducing the temperature components of magnetomodulation DC convertors errors

Reliability is called not only one of the main characteristics of the magnetic modulation DC converters, but also the ability to correctly perform the required function of the measuring device in the specified operating modes and operating conditions for a certain period of time. One of the main tasks in the development of magnetic modulation DC converters is to reduce the number of inference errors. Magnetic modulation DC converters errors include the main errors which are technological and corrective errors, internal sources of additional errors, external sources of additional errors which can be static errors, dynamic errors.

Clinical biostatistics in the 2020s

Clinical biostatistics in the 2020s

Partially Hidden Markov Chain Multivariate Linear Autoregressive model: inference and forecasting—application to machine health prognostics

Time series subject to regime shifts have attracted much interest in domains such as econometry, finance or meteorology. For discrete-valued regimes, models such as the popular Hidden Markov Chain (HMC) describe time series whose state process is unknown at all time-steps. Sometimes, time series are annotated. Thus, another category of models handles the case with regimes observed at all time-steps. We present a novel model which addresses the intermediate case: (i) state processes associated to such time series are modelled by Partially Hidden Markov Chains (PHMCs); (ii) a multivariate linear autoregressive (MLAR) model drives the dynamics of the time series, within each regime. We describe a variant of the expectation maximization (EM) algorithm devoted to PHMC-MLAR model learning. We propose a hidden state inference procedure and a forecasting function adapted to the semi-supervised framework. We first assess inference and prediction performances, and analyze EM convergence times for PHMC-MLAR, using simulated data. We show the benefits of using partially observed states as well as a fully labelled scheme with unreliable labels, to decrease EM convergence times. We highlight the robustness of PHMC-MLAR to labelling errors in inference and prediction tasks. Finally, using turbofan engine data from a NASA repository, we show that PHMC-MLAR outperforms or largely outperforms other models: PHMC and MSAR (Markov Switching AutoRegressive model) for the feature prediction task, PHMC and five out of six recent state-of-the-art methods for the prediction of machine useful remaining life.

Efficient analysis of split-plot experimental designs using model averaging

Split-plot experimental data are often analyzed as if the data came from a completely randomized design. As is well known, ignoring the different levels of randomization and replication can lead to serious inferential errors. However, in some experiments, including many of the ocean global change experiments that motivated this research, variation between whole-plot experimental units may be small relative to variation between subplot units. Even though a factorial analysis will often perform poorly in general, in this special case it outperforms a split-plot analysis, providing narrower confidence intervals for treatment means and differences with coverage rates close to the desired level. The performance of the proposed model-averaged analysis was compared to a classical split-plot analysis via a simulation study, and its utility demonstrated for an ocean global change experiment examining growth and condition of a juvenile mussel species. In our simulation study, model-averaged confidence intervals for whole-plot treatment means or comparisons of means were up to 40% narrower than split-plot confidence intervals while maintaining close to nominal coverage rates. In our example experiment, we observed narrowing of up to 25%. We recommend model averaging as a preferred approach when variation between whole-plot experimental units is expected to be less than between subplot units, with a few caveats for studies with very few replicates.

What have we learned about health effects more than 40 years after the Three Mile Island nuclear accident? A scoping and process review

AbstractThe worst commercial nuclear accident and the first large‐scale voluntary evacuation due to a nuclear event in United States history took place at Three Mile Island (TMI) in 1979. Within a short time, there was a well‐recognized impact of the TMI accident by public health officials and academicians upon mental health of nearby residents—particularly pregnant women. Despite long‐term and detailed follow‐up of this population for cancer outcomes, community concerns continue. Herein, we conduct a scoping and process review of fetal/maternal/child health (MCH) and cancer health outcomes, and consider the findings in light of subsequent accidents. The process evaluation is enhanced by unpublished transcripts of the Pennsylvania Department of Health's (PA DOH) TMI Advisory Panel on Health Research Studies (APHRS), comprised of experts from prominent US research institutions. Research and process strengths included a rapid initial house‐to‐house health census of over 35,000 individuals, intense medical record reviews, in‐person interviews of pregnant women within a 10‐mile radius, and coordination of multiple parallel studies by PA DOH with APHRS collaboration. Major limitations include (1) errors in causal inference, (2) limited radiation exposure assessment, (3) failure to include TMI never‐exposed groups in risk estimates, (4) exclusion of vulnerable populations, (5) nonindexed/never published investigations, and (6) government‐stated need for long‐term MCH follow‐up that was never conducted. Lingering unaddressed actual and perceived risks were identified as major common themes among TMI, Chernobyl and Fukushima accidents. These findings underscore the need for the establishment of mechanisms to ensure long‐term follow‐up and preaccident protocols designed to address major health issues with inclusion of vulnerable populations and perceived risks.

MOHAQ: Multi-Objective Hardware-Aware Quantization of recurrent neural networks

The compression of deep learning models is of fundamental importance in deploying such models to edge devices. The selection of compression parameters can be automated to meet changes in the hardware platform and application. This article introduces a Multi-Objective Hardware-Aware Quantization (MOHAQ) method, which considers hardware performance and inference error as objectives for mixed-precision quantization. The proposed method feasibly evaluates candidate solutions in a large search space by relying on two steps. First, post-training quantization is applied for fast solution evaluation (inference-only search). Second, we propose the ”beacon-based search” to retrain selected solutions only and use them as beacons to estimate the effect of retraining on other solutions. We use speech recognition models on TIMIT dataset. Experimental evaluations show that Simple Recurrent Unit (SRU)-based models can be compressed up to 8x by post-training quantization without any significant error increase. On SiLago, we found solutions that achieve 97% and 86% of the maximum possible speedup and energy saving, with a minor increase in error on an SRU-based model. On Bitfusion, the beacon-based search reduced the error gain of the inference-only search on SRU-based models and Light Gated Recurrent Unit (LiGRU)-based model by up to 4.9 and 3.9 percentage points, respectively.

From Network Inference Errors to Utility Suboptimality: How Much Is the Impact?

Accurate and timely network inference, known as network tomography, is a vital ingredient in efficient network operation, supporting more sophisticated and ambitious traffic optimization algorithms. In practice, however, precise knowledge of the dynamic internal network bottlenecks/states can be impossible to obtain under restricted network visibility on network edge where only end-to-end measurements are available without cooperation of internal network components. In this paper, we acknowledge that network tomography on the edge is often imperfect, leading to dynamic, time-varying inference errors that could change from iteration to iteration on the same time-scale as distributed network optimization algorithms. We quantify the impact of such imperfect bottleneck/state inference on algorithm convergence and optimality. In particular, we show that under arbitrary, bounded inference errors (belonging to three common classes including absent and incorrect bottlenecks, and inaccurate capacity), the solution of the distributed optimization algorithm still converges to a bounded neighborhood of the optimal solution. The resulted optimality gap is quantified in closed form and shown to be proportional to average inference errors. These results are evaluated using extensive network simulations and on real-world IoT data sets. The work provide a theoretical support for understanding the impact of imperfect inference on distributed network optimization.

Differentially Private Tripartite Intelligent Matching Against Inference Attacks in Ride-Sharing Services

In intelligent transportation systems, the key issue of the Ride-Sharing Service (RSS) is to find proper drivers for the passengers by Intelligent Matching (IM) of two or three objects, including the positions of drivers, the travel information of passengers, and the spots where passengers and drivers meet and separate. Unfortunately, the exposure of travel plans of passengers in the IM process due to inference attacks has raised concerns about the privacy violation. To resist the inference attacks, we propose a Differentially Private Tripartite IM (DPTIM) protocol for RSS. DPTIM is based on the tripartite IM process, which intelligently finds the suitable threshold to filter out the matched objects with satisfaction scores below the threshold, so as to provide the high average satisfaction score of matched passengers. Compared to existing relevant mechanisms, DPTIM is distinguished by the feature that it leverages the inference error and differential privacy techniques to prevent the prior-information-based inference attacks and constrain the posterior information leakage, while providing satisfactory matching results. Furthermore, DPTIM meets the personalized demand of location privacy by using the passenger-specific tolerance estimation on inference errors and the personalized privacy budget. Finally, we implement DPTIM on real-world datasets, and demonstrate the satisfactory performance of DPTIM in terms of the average satisfaction score of passengers, the anti-inference-attack capability, and the passenger-specific privacy requirement.

Inference-Based Quantum Sensing.

In a standard quantum sensing (QS) task one aims at estimating an unknown parameter θ, encoded into an n-qubit probe state, via measurements of the system. The success of this task hinges on the ability to correlate changes in the parameter to changes in the system response R(θ) (i.e., changes in the measurement outcomes). For simple cases the form of R(θ) is known, but the same cannot be said for realistic scenarios, as no general closed-form expression exists. In this Letter, we present an inference-based scheme for QS. We show that, for a general class of unitary families of encoding, R(θ) can be fully characterized by only measuring the system response at 2n+1 parameters. This allows us to infer the value of an unknown parameter given the measured response, as well as to determine the sensitivity of the scheme, which characterizes its overall performance. We show that inference error is, with high probability, smaller than δ, if one measures the system response with a number of shots that scales only as Ω(log^{3}(n)/δ^{2}). Furthermore, the framework presented can be broadly applied as it remains valid for arbitrary probe states and measurement schemes, and, even holds in the presence of quantum noise. We also discuss how to extend our results beyond unitary families. Finally, to showcase our method we implement it for a QS task on real quantum hardware, and in numerical simulations.

Criteria to switch from tabulation to neural networks in computational combustion

Motivated by the need to reduce computational costs, look-up tables are widely used in numerical simulations of laminar and turbulent flames, for the thermodynamics of the mixture, for detailed chemistry, and for turbulent combustion closures. At the same time, there have been many studies where artificial neural networks have been trained to replace the classic tabulation approach, and their performance against tabulation typically evaluated a posteriori. In the majority of applications the focus is on accuracy, and the objective is to obtain the best network structure which minimises the inference error during training. Computational efficiency however is also important, and criteria are needed to decide whether or not it is worthwhile in the first place to employ neural networks at all, and if so what the potential bounds on the computational time and memory gains (if any) over tabulation are. This is examined analytically in this work by developing scaling laws for the computational cost of tabulation and of neural networks including the effect of network structure. The scaling laws are validated using both model test-data but also data based on a canonical problem which involves inferring laminar flame speeds of methane/hydrogen mixtures at off-training conditions. The proposed scaling laws lead naturally to a framework for effective decision-making between adopting look-up tables or neural networks.

One Step Forward From Agassi’s Inquiries on Logic: A Fallibilist Logic for Critical Rationalism

Critical rationalists cannot reconcile their falibilism with the demand of logic for universality. Popper tried, but failed, to achieve universality in logic without proof. Attempts to find a limited approach to logic as ‘logics of’ have failed to find a coherent critical rationalist alternative. Critical rationalists take Tarski’s logic to be the best of logic today. But Tarski renders logic as close to justification, and thereby universality, as possible. A fallibilist version of Tarskian logic can yield a critical rationalist alternative: It provides rules for solving problems in linguistic contexts, but also discovers mistakes by discovering errors in logical inference.

A Vertical Single Transistor Neuron with Core-Shell Dual-Gate for Excitatory-Inhibitory Function and Tunable Firing Threshold Voltage.

A novel inhibitable and firing threshold voltage tunable vertical nanowire (NW) single transistor neuron device with core–shell dual-gate (CSDG) was realized and verified by TCAD simulation. The CSDG NW neuron is enclosed by an independently accessed shell gate and core gate to serve an excitatory–inhibitory transition and a firing threshold voltage adjustment, respectively. By utilizing the shell gate, the firing of specific neuron can be inhibited for winner-takes-all learning. It was confirmed that the independently accessed core gate can be used for adjustment of the firing threshold voltage to compensate random conductance variation before the learning and to fix inference error caused by unwanted synapse conductance change after the learning. This threshold voltage tuning can also be utilized for homeostatic function during the learning process. Furthermore, a myelination function which controls the transmission rate was obtained based on the inherent asymmetry between the source and drain in vertical NW structure. Finally, using the CSDG NW neuron device, a letter recognition test was conducted by SPICE simulation for a system-level validation. This multi-functional neuron device can contribute to construct a high-density monolithic SNN hardware combining with the previously developed vertical synapse MOSFET devices.

Sample Size Calculation and Timing of Dose Selection in a Multiple-Dose Clinical Trial

ABSTRACT In a two-stage design for multiple-dose clinical trial, a dose that has the highest observed response rate is often chosen to carry to the next stage, then to compare against the control group. The observed response rate could overestimate the true response rate of the selected dose. If it is used to compare with the response rate of the control group without any adjustment, the Type I error of this statistical inference will not be controlled. In this article, the Stepwise Overcorrection method, which was shown to control the Type I error, is used to derive a formula to calculate the sample size needed for a two-stage multiple-dose design. A maximum-minimum strategy is proposed to ensure the sample size for each group will provide the desired statistical power of detecting the difference in response rates between the selected dose and the control group, assuming at least one of the experimental groups has the desired treatment effect versus the control group.