Showing papers presented at "Static Analysis Symposium in 2021"

Noncontact Neonatal Respiration Rate Estimation Using Machine Vision

Abstract: Using video data of neonates admitted to the neonatal intensive care unit (NICU) we developed and compared the performance of various techniques for noncontact respiration rate (RR) estimation. Data were collected from an overhead colour and depth (RGB-D) camera, while gold standard physiologic data were captured from the hospital's patient monitor. We developed a deep learning algorithm for automatic detection of the face and chest area of the neonate. We then use this algorithm to identify time periods with low patient motion and to locate regions of interest for RR estimation. We produce a respiration signal by quantifying the chest movement using the raw RGB video, motion-magnified RGB video, and depth video. We compare this to a respiration signal derived from the changes in the green channel of the face. We were able to estimate RR from motion-magnified video and depth video, achieving a mean absolute error of less than 3.5 BPM for 69% and 67% of the time for each stream, respectively. We achieve this result without the need for skin segmentation and can apply our technique to fully clothed neonatal patients. We show that similar performance can be achieved using the depth and colour stream using this technique.

Verifying Low-dimensional Input Neural Networks via Input Quantization

Abstract: Deep neural networks are an attractive tool for compressing the control policy lookup tables in systems such as the Airborne Collision Avoidance System (ACAS). It is vital to ensure the safety of such neural controllers via verification techniques. The problem of analyzing ACAS Xu networks has motivated many successful neural network verifiers. These verifiers typically analyze the internal computation of neural networks to decide whether a property regarding the input/output holds. The intrinsic complexity of neural network computation renders such verifiers slow to run and vulnerable to floating-point error.

A Multilanguage Static Analysis of Python Programs with Native C Extensions

Abstract: Modern programs are increasingly multilanguage, to benefit from each programming language’s advantages and to reuse libraries. For example, developers may want to combine high-level Python code with low-level, performance-oriented C code. In fact, one in five of the 200 most downloaded Python libraries available on GitHub contains C code. Static analyzers tend to focus on a single language and may use stubs to model the behavior of foreign function calls. However, stubs are costly to implement and undermine the soundness of analyzers. In this work, we design a static analyzer by abstract interpretation that can handle Python programs calling C extensions. It analyses directly and fully automatically both the Python and the C source codes. It reports runtime errors that may happen in Python, in C, and at the interface. We implemented our analysis in a modular fashion: it reuses off-the-shelf C and Python analyses written in the same analyzer. This approach allows sharing between abstract domains of different languages. Our analyzer can tackle tests of real-world libraries a few thousand lines of C and Python long in a few minutes.

A Dilated Residual Hierarchically Fashioned Segmentation Framework for Extracting Gleason Tissues and Grading Prostate Cancer from Whole Slide Images

Abstract: Prostate cancer (PCa) is the second deadliest form of cancer in males, and it can be clinically graded by examining the structural representations of Gleason tissues. This paper proposes a new method for segmenting the Gleason tissues (patch-wise) in order to grade PCa from the whole slide images (WSI). Also, the proposed approach encompasses two main contributions: 1) A synergy of hybrid dilation factors and hierarchical decomposition of latent space representation for effective Gleason tissues extraction, and 2) A three-tiered loss function which can penalize different semantic segmentation models for accurately extracting the highly correlated patterns. In addition to this, the proposed framework has been extensively evaluated on a large-scale PCa dataset containing 10,516 whole slide scans (with around 71.7M patches), where it outperforms state-of-the-art schemes by 3.22% (in terms of mean intersection-over-union) for extracting the Gleason tissues and 6.91 % (in terms of F1 score) for grading the progression of PCa.

Pengaruh financial literacy, life style, locus of control dan demografi terhadap perilaku konsumtif generasi milenial di kota subang

Abstract: Histori Artikel : Tgl. Masuk : 23 Januari 2021 Tgl. Diterima : 01 Maret 2021 Tersedia Online : 31 Maret 2021 This study aims to determine the effect of financial Literacy, Life style, Locus of control and demographics on the consumption behavior of millennial generation in the city of Subang. Consumptive behavior is an action that prioritizes desires rather than needs, therefore consumption activities are actions taken to meet needs, if consumption is excessive, a consumptive behavior will occur.This research was conducted to the millennial generation in the city of Subang, aged 20 35 years in 2020. The method used in this study is a quantitative method. The number of samples in this study were 200 people, sampling using nonprobability sampling techniques with a purposive sampling approach. The data analysis technique used in this study is multiple linear regression analysis. The results of this study conclude that simultaneous financial literacy, life style, locus of control, and demographics influence the consumptive behavior of millennials in Subang City, Whereas according to the results of the hypothesis partially financial literacy variables, life style, positive and significant influence on consumptive behavior of millennial generation in Subang, locus of control variables have a negative and significant effect on consumptive behavior of millennial generation in Subang, while demographic variables (gender), and demographic (income) does not affect the consumption behavior of millennial generation in the city of Subang.

SmartHand: Towards Embedded Smart Hands for Prosthetic and Robotic Applications

Abstract: The sophisticated sense of touch of the human hand significantly contributes to our ability to safely, efficiently, and dexterously manipulate arbitrary objects in our environment. Robotic and prosthetic devices lack refined tactile feedback from their end-effectors, leading to counterintuitive and complex control strategies. To address this lack, tactile sensors have been designed and developed, but they are either expensive and not scalable or offer an insufficient spatial and temporal resolution. This paper focuses on overcoming these issues by designing a smart embedded system, called SmartHand, enabling the acquisition and real-time processing of high-resolution tactile information from a hand-shaped multi-sensor array for prosthetic and robotic applications. We acquire a new tactile dataset consisting of 340,000 frames while interacting with 16 objects from everyday life and the empty hand, i.e., a total of 17 classes. The design of the embedded system minimizes response latency in classification, by deploying a small yet accurate convolutional neural network on a high-performance ARM Cortex-M7 microcontroller. Compared to related work, our model requires one order of magnitude less memory and 15.6 x fewer computations while achieving similar inter-session accuracy and up to 98.86% and 99.83% top-1 and top-3 cross-validation accuracy, respectively. Experimental results of the designed prototype show a total power consumption of 505mW and a latency of only 100ms.

Improving Thread-Modular Abstract Interpretation

Abstract: We give thread-modular non-relational value analyses as abstractions of a local trace semantics. The semantics as well as the analyses are formulated by means of global invariants and side-effecting constraint systems. We show that a generalization of the analysis provided by the static analyzer Goblint as well as a natural improvement of Antoine Mine’s approach can be obtained as instances of this general scheme. We show that these two analyses are incomparable w.r.t. precision and provide a refinement which improves on both precision-wise. We also report on a preliminary experimental comparison of the given analyses on a meaningful suite of benchmarks.

FPGA Based Meteorological Monitoring Station

Abstract: In this paper, we propose to implement a meteorological monitoring station using embedded systems. This model is possible thanks to different sensors that enable us to measure several environmental parameters, such as i) relative humidity, ii) average ambient temperature, iii) soil humidity, iv) rain occurrence, and v) light intensity. The proposed system is based on a field-programmable gate array device (FPGA). The proposed design aims at ensuring high-resolution data acquisition and at predicting samples with precision and accuracy in real-time. To present the collected data, we develop also a web application with a simple and friendly user interface.

IoT Multi-Hop Facilities via LoRa Modulation and LoRa WanProtocol within Thin Linear Networks

Abstract: This paper proposes a novel network architecture integrating a multi-hop Long Range (LoRa)-based thin linear network within a LoRa Wide Area Network (LoRaWAN) infrastructure, with the aim of proposing linear distributed measurement systems forwarding their collected data to a LoRaWAN server by means of a hybrid LoRa-LoRaWAN node. Such device is able to collect LoRa packets coming from the linear network and to encapsulate them in LoRaWAN packets transmitted to the remote server by means of standard LoRaWAN Gateways. The operation of the nodes is regulated by an ad-hoc routing protocol which aims at minimizing their active period, in order to reduce their power consumption increasing the overall system lifetime. Similarly, the synchronization of the nodes aims at increasing the robustness of the network reducing at minimum packet losses. The effectiveness of the proposed network architecture in terms of successful packet deliveries and reduction of active time is tested in different configurations, exploiting 2-node, 3-node and 4-node chains as well as adopting increasingly larger cycle periods. Results show that the proposed configuration ensures a noteworthy robustness in terms of packets delivery while maintaining the duty-cycling at levels that may guarantee long life times and autonomous operation to the overall infrastructure.

Experimental Demonstration of a 188 meters Infrastructure-to-Vehicle Visible Light Communications Link in Outdoor Conditions

Abstract: Wireless communication technologies have the potential to significantly contribute toward a safer and more efficient road network. In this area, Visible Light Communications (VLC) are on the way of making the transition from an emerging technology to a confirmed technology. In the upper mentioned context, this paper presents the results of an experimental demonstration of Infrastructure-to- Vehicle VLC link in outdoor conditions. For these field tests, a commercial traffic light has been used as a VLC emitter, whereas a photodiode-based VLC receiver has been used to transform the optical beam into an electrical signal. The experimental results demonstrate a communication range of up to 188 meters at a BER of 10–3, with BERs as low as 10–6 for distances below 170 meters. As far as we know, this is the longest I2V VLC link based on standard road side unit equipment reported. Thus, the 188 m I2V VLC link delivered in this paper provides extremely encouraging evidence concerning the use of the VLC technology in automotive applications.

Tackling Time-Variability in sEMG-based Gesture Recognition with On-Device Incremental Learning and Temporal Convolutional Networks

Abstract: Human-machine interaction is showing promising results for robotic prosthesis control and rehabilitation. In these fields, hand movement recognition via surface electromyographic (sEMG) signals is one of the most promising approaches. However, it still suffers from the issue of sEMG signal's variability over time, which negatively impacts classification robustness. In particular, the non-stationarity of input signals and the surface electrodes' shift can cause up to 30 % degradation in gesture recognition accuracy. This work addresses the temporal variability of the sEMG-based gesture recognition by proposing to train a Temporal Convolutional Network (TCN) incrementally over multiple gesture training sessions. Using incremental learning, we re-train our model on stored latent data spanning multiple sessions. We validate our approach on the UniBo-20-Session dataset, which includes 8 hand gestures from 3 subjects. Our incremental learning framework obtains 18.9% higher accuracy compared to a baseline with a standard single training session. Deploying our TCN on a Parallel, Ultra-Low Power (PULP) microcontroller unit (MCU), GAP8, we achieve an inference latency and energy of 12.9 ms and 0.66 mJ, respectively, with a weight memory footprint of 427 kB and a data memory footprint of 0.5-32 MB.

Fast and Efficient Bit-Level Precision Tuning.

Abstract: In this article, we introduce a new technique for precision tuning. This problem consists of finding the least data types for numerical values such that the result of the computation satisfies some accuracy requirement. State of the art techniques for precision tuning use a trial-and-error approach. They change the data types of some variables of the program and evaluate the accuracy of the result. Depending on what is obtained, they change more or less data types and repeat the process. Our technique is radically different. Based on semantic equations, we generate an Integer Linear Problem (ILP) from the program source code. Basically, this is done by reasoning on the most significant bit and the number of significant bits of the values which are integer quantities. The integer solution to this problem, computed in polynomial time by a classical linear programming solver, gives the optimal data types at the bit level. A finer set of semantic equations is also proposed which does not reduce directly to an ILP problem. So we use policy iteration to find the solution. Both techniques have been implemented and we show that our results encompass the results of state-of-the-art tools.

SmartTag: An Ultra Low Power Asset Tracking and Usage Analysis IoT Device with Embedded ML Capabilities

Abstract: Assessing power tools usage helps to prolong their life cycle, as well as indicate targeted maintenance interventions after a particular series of events, e.g. drops. In this work, we propose a low power multi-sensors hardware-software co-design for extremely long shelf life, and a long operating lifecycle. The designed device is based on a Bluetooth Low Energy (BLE) system on chip (SoC) to exchange data with a gateway. NFC has been chosen to wake up the device without adding any additional power consumption. The system on a chip includes an ARM Cortex-M4F core to further process the information achieving low latency and high energy efficiency. The device hosts a temperature and humidity sensor used to monitor the storage conditions, and an accelerometer is used for condition and activity monitoring. This paper provides a proof-of-concept approach to continuously assess the usage of a power tool and detect potential mis-usages, e.g., drops. The architecture, thought to be flexible, can host both traditional signal processing and novel tiny machine learning workloads, offering a future-proof platform for several application scenarios. Experimental results highlight the advanced processing capabilities at low power consumption enabling a long lifetime of up to 4 years with a small coin battery.

Robust Precision Landing for Autonomous Drones Combining Vision-based and Infrared Sensors

Abstract: One of the challenges in drone-based systems is to support automated landing with a high precision that goes beyond the accuracy of standard off-the-shelf GPS. Various efforts have been made to support this, mainly using vision-based and infrared sensors. However, using a single sensor inevitably introduces a single point of failure. To address this problem, we combine a vision-based sensor that detects special visual markers with a sensor that tracks an infrared beacon. We also support a more cautious landing approach for the case where these sensors temporarily fail to detect their targets. We implement our solution in the context of a mature autopilot framework, through modular extensions that are transparent to the rest of the software stack. We evaluate these mechanisms by conducting field experiments using a custom drone, activating faults in the individual precision landing sensor subsystems in a controlled way through interactive commands that are sent to the drone at runtime. The results show that our solution achieves robust precision landing under different failure scenarios while maintaining the accuracy of fault-free sensor operation.

Low-Latency Detection of Epileptic Seizures from iEEG with Temporal Convolutional Networks on a Low-Power Parallel MCU

Abstract: Epilepsy is a severe neurological disorder that affects about 1 % of the world population, and one-third of cases are drug-resistant. Apart from surgery, drug-resistant patients can benefit from closed-loop brain stimulation, eliminating or mitigating the epileptic symptoms. For the closed-loop to be accurate and safe, it is paramount to couple stimulation with a detection system able to recognize seizure onset with high sensitivity and specificity and short latency, while meeting the strict computation and energy constraints of always-on realtime monitoring platforms. We propose a novel setup for iEEG-based epilepsy detection, exploiting a Temporal Convolutional Network (TCN) optimized for deployability on low-power edge devices for real-time monitoring. We test our approach on the Short- Term SWEC-ETHZ iEEG Database, containing a total of 100 epileptic seizures from 16 patients (from 2 to 14 per patient) comparing it with the state-of-the-art (SoA) approach, represented by Hyperdimensional Computing (HD). Our TCN attains a detection delay which is 10s better than SoA, without performance drop in sensitivity and specificity. Contrary to previous literature, we also enforce a time-consistent setup, where training seizures always precede testing seizures chronologically. When deployed on a commercial low-power parallel microcontroller unit (MCU), each inference with our model has a latency of only 5.68 ms and an energy cost of only 124.5 μJ if executed on 1 core, and latency 1.46 ms and an energy cost 51.2 μJ if parallelized on 8 cores. These latency and energy consumption, lower than the current SoA, demonstrates the suitability of our solution for real-time long-term embedded epilepsy monitoring.

Detection of False Alarms in the NICU Using Pressure Sensitive Mat

Abstract: In the neonatal intensive care unit (NICU), a large proportion of alarms are false. This can result in alarm fatigue which increases the risk that alarms of clinical significance are overlooked and may lead to an increased response time. It is therefore of interest to minimize false alarms in the NICU to reduce alarm fatigue. Previous alarm classification systems rely on physiologic data and waveforms. In this study, we explore the use of a pressure sensitive mat (PSM), which is an unobtrusive and non-contact secondary sensor system that captures motion-related data. We use a dataset of 136 manually annotated alarm events for 10 neonatal subjects to train a machine learning model for the detection of false alarms. Results show that a combination of physiologic and PSM features has the best performance, which achieves a 0.87 macro-averaged F1 score, compared to the model that solely relies on physiologic data which only achieves a 0.73 macro-averaged F1 score. We also show that the use of PSM data improves the model's ability to generalize to unseen patients using a leave-one-subject-out test protocol. This study demonstrates that the PSM provides complementary and useful information for Improving the discrimination of true and false alarms.

Selective Context-Sensitivity for k-CFA with CFL-Reachability

Abstract: k-CFA provides the most well-known context abstraction for program analysis, especially pointer analysis, for a wide range of programming languages. However, its inherent context explosion problem has hindered its applicability. To mitigate this problem, selective context-sensitivity is promising as context-sensitivity is applied only selectively to some parts of the program. This paper introduces a new approach to selective context-sensitivity for supporting k-CFA-based pointer analysis, based on CFL-reachability. Our approach can make k-CFA-based pointer analysis run significantly faster while losing little precision, based on an evaluation using a set of 11 popular Java benchmarks and applications.

LiDAR + Camera Sensor Data Fusion On Mobiles With AI-based Virtual Sensors To Provide Situational Awareness For The Visually Impaired

Abstract: Autonomy of the blind and visually impaired can be achieved through technological means and thereby empowering them with a sense of independence. Mobile phones are ubiquitous and can access artificial intelligence capabilities locally and in the Cloud. Navigational sensors, such as Light Detection and Ranging (LiDAR), and wide angle cameras, typically found in self-driving cars, are beginning to be incorporated into mobile phones. In this paper, we propose techniques for using mobile phone LiDAR + camera sensor data fusion along with edge + Cloud split AI to create an indoor situational awareness and navigational aid for the visually impaired. In addition to physical sensors, the system uses AI models as virtual sensors to provide the required functionality. The system enhances the image of a scene captured by a camera using distance information from the LiDAR and directional information computed by the device to provide a rich 3-D description of the space in front of the user. The system also uses a combination of sensor data fusion and geometric formulas to provide step-by-step walking instructions for the user in order to reach destinations. The user-centric system proposed here can be a valuable assistive technology for the blind and visually imnpired.

Novel Method of Temperature Modulation for Enhancing Catalytic Gas Sensor Selectivity

Abstract: Catalytic gas sensors are among the most widespread gas sensors for combustible gas concentration measurements. However, their selectivity is low. In this research, the results of machine learning techniques application to enhance catalytic gas sensor selectivity are presented. The measurements of sensor signal are performed using the multistage heat pulse method described in our previous works. Contrary to the previous works, the number of heating stages was increased from 2 to 55, which corresponds to the heating voltage range of 125 m V to 1.5 V with a 25 m V step. This change enriches sensor signal with information about gas compositions. Methane and vapors of acetone, ethanol and gasoline are used as target gases. A support vector machine method is used to train two models. The first one was trained based on the plain normalized data. It was used for a microcontroller implementation of the method. The second model used the data transformed by principal component analysis technique. This model was used to visualize the method proposed. The results show that the application of proposed method allows to identify gases by single catalytic sensor. These principles can be used to design selective gas detectors which will react only to target gases.

Polycrystalline silicon photovoltaic harvesting for indoor IoT systems under red- far red artificial light

Abstract: This paper aims at demonstrating the feasibility of a LoRaWAN-based sensor node for temperature monitoring, autonomously powered by a polycrystalline silicon photovoltaic module with possible applications within the Internet of Things (loT) domain in the horticulture field. The commercial solar cell was characterized under two light sources: a conventional white 4000 K Light Emitting Diode (LED) and a red and far red (R:FR) lamp peaked at 655 nm and 730 nm. The sensor node is equipped with a RFM95x LoRa transceiver which proved to be a valid technology in those application scenarios where robustness and low power consumption are required. The energy harvesting features are performed by a nano-power boost charger buck converter which deals with the power extraction from the photovoltaic module, the LiPo battery charging/discharging management and the supply of the sensor node. Field tests demonstrate that under R:fr light source, the energy self-sufficiency of the system is achieved: a positive balance between the battery charge and discharge is measured, sufficient both for the node working operation and for the battery charging.

RGB-D Sensor Application for Non-Contact Neonatal Monitoring

Abstract: RGB-D cameras have shown promise in noncontact monitoring of patients in the neonatal intensive care unit (NICU). This work conducts essential experiments to assess the suitability and safe use of the Intel RealSense SR300 camera for non-contact neonatal monitoring. Since a pulse oximeter monitoring the patient's oxygen saturation levels (SpO2) senses infrared light, and the RGB-D sensor has an infrared projector, this work investigates a safe camera distance to ensure that the projected infrared light from the camera does not interfere with the SpO2 signal. RGB-D data reflection artifacts from the Plexiglass surface are also explored for single- and double-walled incubators. To prevent from infrared interference and RGB-D data artifacts, we recommend placing the camera at a minimum distance of 40 cm for open beds, and 25 cm for closed incubators. The camera should also be mounted directly on the Plexiglass surface in closed incubators, especially for double-wall designs. We have developed a custom latex apparatus to adhere an SR300 camera to the outer surface of an incubator to avoid reflections while securely mounting the camera without requiring any modification to the incubator itself. This work provides critical information for safe and practical RGB-D camera application in non-contact neonatal monitoring.

Partial Discharge Detection Using Distributed Acoustic Sensing at the Oil-Pressboard Interface

Abstract: This paper investigates novel, initial experimentation in detecting and analysing Partial Discharge at the Oil-Pressboard interface using a continuous fibre-optic-based Distributed Acoustic Sensing (DAS) system. Discharge was successfully detected at a minimum of 223 pC despite the sample rate of DAS being lower than the spectra of acoustic emission. DAS presents multiple advantages over conventional Partial Discharge techniques including inherent localisation, immunity to electrical and magnetic noise, as well as much greater detection distances.

Reduced Products of Abstract Domains for Fairness Certification of Neural Networks.

Abstract: We present Libra, an open-source abstract interpretation-based static analyzer for certifying fairness of ReLU neural network classifiers for tabular data. Libra combines a sound forward pre-analysis with an exact backward analysis that leverages the polyhedra abstract domain to provide definite fairness guarantees when possible, and to otherwise quantify and describe the biased input space regions. The analysis is configurable in terms of scalability and precision. We equipped Libra with new abstract domains to use in the pre-analysis, including a generic reduced product domain construction, as well as search heuristics to find the best analysis configuration. We additionally set up the backward analysis to allow further parallelization. Our experimental evaluation demonstrates the effectiveness of the approach on neural networks trained on a popular dataset in the fairness literature.

Static Analysis of ReLU Neural Networks with Tropical Polyhedra

Abstract: This paper studies the problem of range analysis for feedforward neural networks, which is a basic primitive for applications such as robustness of neural networks, compliance to specifications and reachability analysis of neural-network feedback systems. Our approach focuses on ReLU (rectified linear unit) feedforward neural nets that present specific difficulties: approaches that exploit derivatives do not apply in general, the number of patterns of neuron activations can be quite large even for small networks, and convex approximations are generally too coarse. In this paper, we employ set-based methods and abstract interpretation that have been very successful in coping with similar difficulties in classical program verification. We present an approach that abstracts ReLU feedforward neural networks using tropical polyhedra. We show that tropical polyhedra can efficiently abstract ReLU activation function, while being able to control the loss of precision due to linear computations. We show how the connection between ReLU networks and tropical rational functions can provide approaches for range analysis of ReLU neural networks. We report on a preliminary evaluation of our approach using a prototype implementation.

3D printed capacitive shear and normal force sensor using a highly flexible dielectric

Abstract: We have investigated an entirely 3D printed capacitive sensor. Using a combination of 4 variable capacitors it allows to simultaneously measure shear and normal forces. To guide the design and analysis the behavior of the sensor has been modeled using both finite element method (FEM) simulations and an analytical model. The sensor was tested in a mechanical test setup by means of a linear actuator, loading the sensor with a force from various angles. The sensor showed it was able to measure both the normal and shear force components with a maximum noise floor of 1.5 N.

Quasi-Static Magnetic Localization of Capsule Endoscopes with an Active Integrated Coil

Abstract: The reliable localization for capsule endoscopes is an open research topic. In this study, a low-frequency magnetic localization method for capsule endoscopes with an integrated active coil is proposed. The spatial constraint, the limited battery capacity and the ferromagnetic battery shell of commercial capsules were considered. The generated magnetic flux density was evaluated depending on the distance to the coil and the maximal detectable range was determined. Twelve sensors were arranged in rings and by comparing the measured magnetic flux density with the analytic dipole model, the position and orientation of the coil were reconstructed. The results revealed that the ferromagnetic shell increases the magnetic moment of the coil by approximately a factor of 2.4. Moreover, the mean position and orientation errors were 0.5 mm and 0.3°. Furthermore, by using only the three closest sensors to the coil, a similar localization performance was achieved. Therefore, it was concluded that it is a feasible approach to choose the sensors, which measure the strongest signal to address the problem of the maximal detectable range of magnetic sensors. Moreover, by considering the limited battery capacity, the localization with the proposed coil must be conducted in short time intervals instead of continuously.

Preliminary Results for the Automated Assessment of Driving Simulation Results for Drivers with Cognitive Decline

Abstract: Aging related changes and pathology affecting cognition and the ability to drive are significant issues for individuals, their families and the general population. Ensuring that unsafe drivers have their license suspended or get the additional training they need is important for the safety of the general population. On the other hand, allowing a person to continue to drive as long as they are safe is important for the social, emotional and cognitive wellbeing of the individual. This paper presents results of a preliminary study to see if an automated assessment based on trained machine learning models can correctly classify simulator drives as safe or unsafe in comparison to expert driver assessment opinion. The results show that the machine learning is able to achieve 85% accuracy in comparison to the experts for a combined group of 47 drivers that included 20 Healthy Controls, 9 diagnosed with Lewy Body Dementia and 18 diagnosed with mild Dementia of Alzheimer's Type. This work shows the potential for automated driver simulation assessment, which could reduce the burden on clinicians regarding driver safety evaluation.

Backward Symbolic Execution with Loop Folding

Abstract: Symbolic execution is an established program analysis technique that aims to search all possible execution paths of the given program. Due to the so-called path explosion problem, symbolic execution is usually unable to analyze all execution paths and thus it is not convenient for program verification as a standalone method. This paper focuses on backward symbolic execution (BSE), which searches program paths backwards from the error location whose reachability should be proven or refuted. We show that this technique is equivalent to performing k-induction on control-flow paths. While standard BSE simply unwinds all program loops, we present an extension called loop folding that aims to derive loop invariants during BSE that are sufficient to prove the unreachability of the error location. The resulting technique is called backward symbolic execution with loop folding (BSELF). Our experiments show that BSELF performs better than BSE and other tools based on k-induction when non-trivial benchmarks are considered. Moreover, a sequential combination of symbolic execution and BSELF achieved very competitive results compared to state-of-the-art verification tools.

Toward Neural-Network-Guided Program Synthesis and Verification

Abstract: We propose a novel framework of program and invariant synthesis called neural network-guided synthesis. We first show that, by suitably designing and training neural networks, we can extract logical formulas over integers from the weights and biases of the trained neural networks. Based on the idea, we have implemented a tool to synthesize formulas from positive/negative examples and implication constraints, and obtained promising experimental results. We also discuss an application of our method for improving the qualifier discovery in the framework of ICE-learning-based CHC solving, which can in turn be applied to program verification and inductive invariant synthesis. Another potential application is to a neural-network-guided variation of Solar-Lezama’s program synthesis by sketching.

A novel approach for human activity recognition using object interactions and machine learning

Abstract: Recognising human activity can be advantageous in a number of different scenarios including elder care, healthcare or for training purposes. It can be of direct use to support humans in doing different activities, but is still a challenge for systems to correctly classify the activity in a way that is valuable for the user, as they often times lack the robustness or simplicity for day-to-day use. In this paper an approach for human activity recognition based on object interactions is presented. The proposed system consists of a wireless sensor network, with each sensor node measuring the received signal strength indication (RSSI) to its neighbouring nodes. The accumulated RSSI data is then analyzed by a machine learning algorithm which tries to infer one of several cooked dishes from that data. Experimental studies demonstrate promising results and therefore potential for this technology for recognising human activity in the form of cooking, but its generalised approach makes it suitable for other environments, too.