Behavior modification | EPFL Graph Search

Behavior modification is an early approach that used respondent and operant conditioning to change behavior. Based on methodological behaviorism, overt behavior was modified with consequences, including positive and negative reinforcement contingencies to increase desirable behavior, or administering positive and negative punishment and/or extinction to reduce problematic behavior. It also used Flooding desensitization to combat phobias. Applied behavior analysis (ABA)—the application of behavior analysis—is the current term and is based on radical behaviorism, which refers to B. F. Skinner's viewpoint that cognition and emotions are covert behavior that are to be subjected to the same conditions as overt behavior. Description The first use of the term behavior modification appears to have been by Edward Thorndike in 1911. His article Provisional Laws of Acquired Behavior or Learning makes frequent use of the term "modifying behavior". Through early research in the 1940s and

Subquadratic Overparameterization for Shallow Neural Networks

Volkan Cevher, Armin Eftekhari, Thomas Michaelsen Pethick, Chaehwan Song

Overparameterization refers to the important phenomenon where the width of a neural network is chosen such that learning algorithms can provably attain zero loss in nonconvex training. The existing theory establishes such global convergence using various initialization strategies, training modifications, and width scalings. In particular, the state-of-the-art results require the width to scale quadratically with the number of training data under standard initialization strategies used in practice for best generalization performance. In contrast, the most recent results obtain linear scaling either with requiring initializations that lead to the “lazy-training”, or training only a single layer. In this work, we provide an analytical framework that allows us to adopt standard initialization strategies, possibly avoid lazy training, and train all layers simultaneously in basic shallow neural networks while attaining a desirable subquadratic scaling on the network width. We achieve the desiderata via Polyak-Łojasiewicz condition, smoothness, and standard assumptions on data, and use tools from random matrix theory.

2021

Generic DSP-based Implantable Body Sensor Node

Léandre Bolomey

To improve health, to help to manage chronical disease and to provide equal access to healthcare, continuous monitoring of physiological parameter associated to telemedicine is mandatory. In this perspective Body Sensor Networks concept has emerged. It includes wearable and implantable body sensors with specific focuses on ultra-low power processing and communication, power scavenging, autonomous sensing, data mining, intelligent on-node processing and integrated wireless sensor microsystems. These aim to detect the early disease symptoms in at risk group such as the elderly or assisted patient with chronic illness and to provide them with accurate therapy reducing associated complications. The primary motivation of implantable Body Sensor Node (BSN) development is to provide long-term continuous monitoring without activity restriction, behavior modification and lower cost of care delivery. In this thesis, we propose and assess a minimally invasive BSN that can fulfill the requirements of the widespread range of emerging biosensor and bioactuator. The node architecture is based on a Digital Signal Processor (DSP) combined to a wireless communication transceiver compliant to the Medical Implant Communication Service (MICS) standard. The node exhibits ultra-low power consumption, high computation capabilities and high data rate transmission thanks to the implementation of state of the art integrated circuits. Original packaging methods have been developed to optimize volume occupancy of the main BSN components i.e. integrated circuits, antenna, power source and biosensors. The electronic components have been packaged on a flexible System in Package (SiP) folded in accordion together with chip-on-flip-chip assembly to obtain the highest density in a minimal volume. Jointly with the LEMA at the EPFL, an electrical small dual-band antenna has been developed to achieve far field communication. In addition to its minimally invasive size, the cylindrical BSN packaging is also adaptable. It allows modification in power source and biosensor or bioactuator volume by varying its length depending on the power consumption of the application. The performances of the generic BSN are assessed through a glucose biosensor based on a microviscometer. This application demonstrates the capacity of the system to deliver peak power as high as 80 mW for actuation. Additionally, the efficient computation ability of the architecture and its ultra-low power are confirmed with an average power of only 20 µW. In parallel the BSN radio frequency benchmarking in-vitro revealed a communication distance as far as 14 meters. With the synergy of the LDCS at the EPFL, in-vivo BSN assessment has been possible through a skin temperature measurement experiment. Two BSN have been implanted subcutaneously in a small farm pig. During 15 days the BSN recorded continuously the temperature at two different depths under the pig skin. To these data we also added an external temperature sensor and a surveillance camera to follow the pig behavior in its stall. Thus we built a heterogeneous network of implantable and external sensors that could be remotely accessed. With this BSN development, we paved the way for implantable body sensor network and consequently for future "telehealth" or telemedicine applications. The generic feature of the BSN enables the implementation of other kinds of biosensors or therapeutic devices such as pH, glucose, blood pressure, heart rhythm and the forthcoming one.

EPFL2010

Generic DSP-based Implantable Body Sensor Node

Léandre Bolomey

EPFL2010