Scalable VLSI implementations for neural networks

This paper discusses research on scalable VLSI implementations of feed-forward and recurrent neural networks. These two families of networks are useful in a wide variety of important applications—classification tasks for feed-forward nets and optimization problems for recurrent nets—but their differences affect the way they should be built. We find that analog computation with digitally programmable weights works best for feed-forward networks, while stochastic processing takes advantage of the integrative nature of recurrent networks. We have shown early prototypes of these networks which compute at rates of 1–2 billion connections per second. These general-purpose neural building blocks can be coupled with an overall data transmission framework that is electronically reconfigured in a local manner to produce arbitrarily large, fault-tolerant networks.     

On thermodynamics and gravity

Physics of Atomic Nuclei - A new entropic gravity inspired derivation of general relativity from thermodynamics is presented. This generalizes the “Thermodynamics of Spacetime” approach...     

A comparison of the noise sensitivity of nine QRS detectionalgorithms

The noise sensitivities for nine different QRS detection algorithms were measured for a normal, single-channel lead II, synthesized ECG corrupted with five different types of synthesized noise. The noise types were electromyographic interference, 60 Hz powerline interference, baseline drift due to respiration, abrupt baseline shift, and a composite noise constructed from all of the other noise types. The percentage of QRS complexes detected, the number of false positives, and the detection delay were measured. None of the algorithms were able to detect all QRS complexes without any false positives for all of the noise types at the highest noise level. Algorithms based on amplitude and slope had the highest performance for EMG-corrupted ECG. An algorithm using a digital filter had the best performance for the composite noise corrupted data.     

Emission and capture of electrons in multiquantum-well structures

The mechanisms of unipolar emission and capture of electrons are studied in multiquantum-well structures in relation with the quantum-well infrared photoconductors (QWIP's). We clarify the roles played by the physical parameters which appear in the different QWIP photoresponse models, i.e., the photoconductive and the photoemissive ones. We then describe the experimental procedures which allow us to independently determine these different parameters: deep level optical spectroscopy for the electron emission probability, impedance spectroscopy for the quantum-well capture velocity and thermal emission time, as well as the infrared photoconductive gain for the unipolar electron capture time. The measured dependence of these parameters on photon energy and electric field sheds light on the microscopic physical phenomena which are involved in quantum-well infrared photodetection. Theoretical results on optical phonon mediated transitions in an applied electric field from barrier to well states show good agreement with experiment at low fields but less dependence on the field. Finally, this theoretical approach allows us to connect the different parameters and solve the apparent discrepancy between the QWIP photoresponse models     

Temperature dependence of common-emitter I-V andcollector breakdown voltage characteristics in AlGaAs/GaAs andAlInAs/GaInAs HBT's grown by MBE

Temperature-dependent measurements from 25 to 125°C have been made of the DC I-V characteristics of HBTs with GaAs and In_0.53Ga_0.47As collector regions. It was found that the GaAs HBTs have very low output conductance and high collector breakdown voltage BV_CEO>10 V at 25°C, which increases with temperature. In striking contrast, the In_0.53Ga_0.47As HBTs have very high output conductance and low BV_CEO~2.5 V at 25°C, which actually decreases with temperature. This different behavior is explained by the >10⁴ higher collector leakage current, I_CO, in In_0.53Ga_0.47As compared to GaAs due to bandgap differences. It is also shown that device self-heating plays a role in the I-V characteristics     

The notion of motion: covariational reasoning and the limit concept

This paper investigates outcomes of building students’ intuitive understanding of a limit as a function's predicted value by examining introductory calculus students’ conceptions of limit both before and after instruction. Students’ responses suggest that while this approach is successful at reducing the common limit equals function value misconception of a limit, new misconceptions emerged in students’ responses. Analysis of students’ reasoning indicates a lack of covariational reasoning that coordinates changes in both x and y may be at the root of the emerging limit reached near x = c misconception. These results suggest that although dynamic interpretations of limit may be intuitive for many students, care must be taken to foster a dynamic conception that is both useful at the introductory calculus level and is in line with the formal notion of limit learned in advanced mathematics. In light of the findings, suggestions for adapting the pedagogical approach used in this study are provided.     

On Packet Switches with Infinite Storage

Most prior work on congestion in datagram systems focuses on buffer management. We find it illuminating to consider the case of a packet switch with infinite storage. Such a packet switch can never run out of buffers. It can, however, still become congested. The meaning of congestion in an infinite-storage system is explored. We demonstrate the unexpected result that a datagram network with infinite storage, first-in, first-out queueing, at least two packet switches, and a finite packet lifetime will, under overload, drop all packets. By attacking the problem of congestion for the infinite-storage case, we discover new solutions applicable to switches with finite storage.     

Onsager,ice, biomembranes,dimer models and the F-model

Personal recollections give a sampling of some of Onsager's later interests in ice and biomembranes. The author's involvement in these topics led to modeling a particular biomembrane phase transition using dimer models. Recent work is described for a particularly rich dimer model which is isomorphic to the F-model in three kinds of fields (direct, staggered, and quarter). New results for the full three-dimensional phase diagram show (1) how the anomalous OK multicritical point is destroyed by a direct field, and (2) how a new line of critical points must be added to the phase diagram in direct and staggered fields obtained previously by Baxter, due to diverging susceptibilities in the quarter field.     

Efficacy of hydrolytic enzyme augmentation and thermochemical pretreatments for increased secondary anaerobic digestion of treated municipal sewage sludges

Rising costs for landfill disposal of municipal sewage residues have prompted evaluation of alternative methods for reducing the bulk of the final waste. Representative samples of municipal sewage sludge residues were obtained from three major treatment plants in the United States, including Los Angeles (Hyperion), Denver (North Metro), and Chicago (Stickney). The majority of the treated, dewatered sewage sludge solids was found to be volatile (50–60%) and, presumably, biodegradable. Additionally, much of the volatile content was solubilized by both acid detergent fiber and neutral detergent fiber treatments, and was presumed to be proteineous microbial biomass in nature. Both low- and high-solids anaerobic digester systems, as well as the standard biochemical methane potential (BMP) assay, were utilized to evaluate the anaerobic digestibility of these sewage sludge residues. The low methane yields and, thus, the poor organic waste conversion indicated the need for treatment prior to bioconversion. The effectivenesss of various pretreatments based on assessment of increased soluble protein or organics and anaerobic digestibility as determined by the BMP assay was evaluated.