VLSI implementation of neural networks

Currently, fuzzy controllers are the most popular choice for hardware implementation of complex control surfaces because they are easy to design. Neural controllers are more complex and hard to train, but provide an outstanding control surface with much less error than that of a fuzzy controller. There are also some problems that have to be solved before the networks can be implemented on VLSI chips. First, an approximation function needs to be developed because CMOS neural networks have an activation function different than any function used in neural network software. Next, this function has to be used to train the network. Finally, the last problem for VLSI designers is the quantization effect caused by discrete values of the channel length (L) and width (W) of MOS transistor geometries. Two neural networks were designed in 1.5 microm technology. Using adequate approximation functions solved the problem of activation function. With this approach, trained networks were characterized by very small errors. Unfortunately, when the weights were quantized, errors were increased by an order of magnitude. However, even though the errors were enlarged, the results obtained from neural network hardware implementations were superior to the results obtained with fuzzy system approach.     

Hardware implementation of a new artificial neuron

This paper describes an FPGA (Field Programmable Gate Arrays) implementation of a new type of neuron, the Quantron. The goal is to demonstrate the capability of current technology to closely recreate the human body's reaction to a change of temperature. This is accomplished by creating a function that adds a number of kernels at different frequencies depending on the external temperature. Once the sum of the kernels reaches a certain threshold, the artificial neural network, equivalent to its biological counterpart, "reacts" by sending a specific output signal designed to trigger a response. The various elements of each subsystem are discussed and implemented in software and hardware. The results are analyzed in terms of accuracy and efficiency compared to the biological equivalent.     

Schema design and implementation of the grasp-related mirror neuron system

 Mirror neurons within a monkey's premotor area F5 fire not only when the monkey performs a certain class of actions but also when the monkey observes another monkey (or the experimenter) perform a similar action. It has thus been argued that these neurons are crucial for understanding of actions by others. We offer the hand-state hypothesis as a new explanation of the evolution of this capability: the basic functionality of the F5 mirror system is to elaborate the appropriate feedback – what we call the hand state– for opposition-space based control of manual grasping of an object. Given this functionality, the social role of the F5 mirror system in understanding the actions of others may be seen as an exaptation gained by generalizing from one's own hand to an other's hand. In other words, mirror neurons first evolved to augment the “canonical” F5 neurons (active during self-movement based on observation of an object) by providing visual feedback on “hand state,” relating the shape of the hand to the shape of the object. We then introduce the MNS1 (mirror neuron system 1) model of F5 and related brain regions. The existing Fagg–Arbib–Rizzolatti–Sakata model represents circuitry for visually guided grasping of objects, linking the anterior intraparietal area (AIP) with F5 canonical neurons. The MNS1 model extends the AIP visual pathway by also modeling pathways, directed toward F5 mirror neurons, which match arm–hand trajectories to the affordances and location of a potential target object. We present the basic schemas for the MNS1 model, then aggregate them into three “grand schemas”– visual analysis of hand state, reach and grasp, and the core mirror circuit – for each of which we present a useful implementation (a non-neural visual processing system, a multijoint 3-D kinematics simulator, and a learning neural network, respectively). With this implementation we show how the mirror system may learnto recognize actions already in the repertoire of the F5 canonical neurons. We show that the connectivity pattern of mirror neuron circuitry can be established through training, and that the resultant network can exhibit a range of novel, physiologically interesting behaviors during the process of action recognition. We train the system on the basis of final grasp but then observe the whole time course of mirror neuron activity, yielding predictions for neurophysiological experiments under conditions of spatial perturbation, altered kinematics, and ambiguous grasp execution which highlight the importance of the timingof mirror neuron activity. Received: 6 August 2001 / Accepted in revised form: 5 February 2002     

An analog VLSI implementation of a visual interneuron: enhanced sensory processing through biophysical modeling

Flies are capable of rapid, coordinated flight through unstructured environments. This flight is guided by visual motion information that is extracted from photoreceptors in a robust manner. One feature of the fly's visual processing that adds to this robustness is the saturation of wide-field motion-sensitive neuron responses with increasing pattern size. This makes the cell's responses less dependent on the sparseness of the optical flow field while retaining motion information. By implementing a compartmental neuronal model in silicon, we add this "gain control" to an existing analog VLSI model of fly vision. This results in enhanced performance in a compact, low-power CMOS motion sensor. Our silicon system also demonstrates that modern, biophysically-detailed models of neural sensory processing systems can be instantiated in VLSI hardware.     

Distinct rhythmic locomotor patterns can be generated by a simple adaptive neural circuit: Biology,simulation, and VLSI implementation

Rhythmic motor patterns can be induced in leg motor neurons of isolated locust thoracic ganglia by bath application of pilocarpine. We observed that the relative phases of levators and depressors differed in the three thoracic ganglia. Assuming that the central pattern generating circuits underlying these three segmental rhythms are probably very similar, we developed a simple model circuit that can produce any one of the three activity patterns and characteristic phase relationships by modifying a single synaptic weight. We show results of a computer simulation of this circuit using the neuronal simulator NeuraLOG/Spike. We built and tested an analog VLSI circuit implementation of this model circuit that exhibits the same range of behaviors as the computer simulation. This multidisciplinary strategy will be useful to explore the dynamics of central pattern generating networks coupled to physical actuators, and ultimately should allow the design of biologically realistic walking robots.     

Artificial neuron

Zusammenfassung Frühere Arbeiten liefen darauf hinaus, ein Nervenelement (das künstliche Neuron) qualitativ zu simulieren. Im folgenden werden die Ergebnisse der weiteren Entwicklungsstufe beschrieben — die Simulierung eines Neuronenanalogons, dessen Zeitbasis mit der der Aplysia identisch ist und dessen Spannungsamplitude(n) sich voneinander um einen Maßstabsfaktor unterscheiden.Dieses künstliche Neuron, das aus einem einzigen Operationsverstärker mit parallelverlaufenden linearen und nichtlinearen Rückkopplungsleitungen besteht, weist nicht nur eine quantitative Beziehung zu dem natürlichen auf; es ist auch in der Arbeitsweise und dem Schaltschema nach viel einfacher. Die Verhaltenscharakteristiken des Analogous, mit bzw. ohne Rückkopplung, werden beschrieben.Einige von den dargestellten und im Text erläuterten Charakteristiken sind: Ansetzen eines Wirkungspotentials durch depolarisierende Eingänge, Anodendurchschlagsspannung, Wiederpolarisation, gegen Totzeit aufgetragene Impulsstärke, Akkommodation, Adaptation und wiederholte Erzeugung des Wirkungspotentials durch andauerndes depolarisierendes Reizen.

---

---

This work was supported by the Bureau of Naval Weapons, Department of the Navy, under Contract NOw 62-0604-c.     

Analog circuit of theAcetabularia membrane

Summary The high membrane potential ofAcetabularia (E _m=–170 mV) is due to an electrogenic pump in parallel with the passive diffusion system (E _d=–80 mV) which could be studied separately in the cold, when the pump is blocked. Electrical measurements under normal conditions show that the pump pathway consists of its electromotive forceE _p with two elementsP ₁ andP ₂ in series;P ₂ is shunted by a large capacitance (C _p=3 mF cm^–2). The nonlinear current-voltage relationship ofP ₁ (light- and temperature-sensitive) could be determined separately; it reflects the properties of a carrier-mediated electrogenic pump. The value ofE _p (–190 mV) indicates a stoichiometry of 21 between electrogenically transported charges and ATP. The electrical energy, normally stored inC _p, compares well with the metabolic energy, stored in the ATP pool. The nonlinear current-voltage relationship ofP ₂ (attributed to phosphorylating reactions) is also sensitive to light and temperature and is responsible for the region of negative conductance of the overall current-voltage relationship. The power of the pump (1 W cm^–2) amounts to some percent of the total energy turnover. The high Cl^– fluxes (1 nmol cm^–2 sec^–1) and the electrical properties of the plasmalemma are not as closely related as assumed previously. For kinetic reasons, a direct and specific Cl^– pathway between the vacuole and outside is postulated to exist.     

Analog modelling of cochlear adaptation

Adaptation phenomena in the peripheral hearing organ originate from the properties of the haircell-neuron synaps as revealed by their temperature dependency. In order to verify this hypothesis a model experiment is set up. The model consists of programming on an analog computer the equations describing the reaction kinetics from the frogs myoneural junction. A model neuron is attached to the synapsmodel in which two independent noise sources are incorporated. This noise addition serves to simulate the stochastic behavior of the transmitter release in the synaps resulting in a fluctuating generator potential and independently thereof to reflect the varying threshold of the nerve fiber. The reaction rate constants in the synaptic model were modified with respect to the original ones in order to get a coincidence of in vivo- and model results. The compound model primarily is used to simulate the quite different synchronization between the auditory nerve fibers occurring during intensity changes and during changes of the stimulus repetition rate. These results were known to be different in the animal experiments and were also generated by the model. It is shown clearly that neither synaptic noise alone nor membrane noise alone can account for the observations made in the animal experiments. Therefore a combination of both types of noise is used in this model. The model experiment also visualizes that amplitude changes for compound AP's during adaptation can be explained in toto by a decreasing synchronization, i.e. the broadening of the latency distribution function for the nerve fibers.     

Emergence of neuron types

Neuron types are the building blocks of the nervous system, and therefore, of functional circuits. Understanding the origin of neuronal diversity has always been an essential question in neuroscience and developmental biology. While knowledge on the molecular control of their diversification has largely increased during the last decades, it is now possible to reveal the dynamic mechanisms and the actual stepwise molecular changes occurring at single-cell level with the advent of single-cell omics technologies and analysis with high temporal resolution. Here, we focus on recent advances in the field and in technical and analytical tools that enable detailed insights into the emergence of neuron types in the central and peripheral nervous systems.     

Selectivity of chemoreceptor neuron

Discriminating ability (selectivity) of chemoreceptor neuron is compared with that of its receptor proteins. The process of neuronal triggering is expected to be cooperative and threshold-type in a sense that the neuron can fire if and only if the number of its receptor proteins, which are bound with odor molecules, is above a definite threshold. Both deterministic and stochastic pictures are considered. The stochastic case is treated based on birth and death stochastic process and first passage technique. In both pictures, it is shown that a chemoreceptor neuron can have much a higher selectivity than its individual receptor proteins, provided the chemical stimuli are presented at low concentrations, and the threshold is high enough. This is in agreement with a preliminary estimate based on simplified probabilistic reasoning (Vidybida, A.K., 1999. Cooperative mechanism for improving the discriminating ability in the chemoreceptive neuron. Binomial case. Biol. Cybern. 81, 469-473). The mechanism of selectivity improvement is similar to that described before in cooperative chemical systems. A possibility for this mechanism to be valid at higher stages of processing of chemical signals, as well as in other sensory systems is discussed.     

The prospects for analogue neural VLSI

In recent years, the efforts of analogue, neural-hardware designers have shifted from generic analogue neurocomputers to "niche" markets in sensor fusion and robotics, and we explain why this is so. We describe the main differences between digital and analogue computation, and consider the advantages of pure analogue and pulsed methods of design. We then investigate some important issues in analogue design of neural machines, namely weight storage (volatile and non-volatile), on-chip learning, and arithmetic accuracy and its relationship to noise. Finally, we outline those areas in which analogue techniques are likely to prove most useful, and speculate as to their likely long-term utility.     

Low-Frequency Vibrations of a Nucleoside Analog

Abstract

This paper reports on the first coherent neutron scattering measurements ever carried out on the vibrational spectrum of a nucleoside analog. Frequencies up to 3.5 Thz belonging to some half-dozen dispersion curves have been measured in a plane normal to the [110] direction in a triclinic crystal of 5-iodo-2′-deoxyuridine (5-iodouridine). The number of phonon branches observed suggests that the deoxyribose sugar and pyrimidine in this molecule are the vibrating entities at low frequencies. Comparisons, notably of elastic properties, are made with previous measurements on crystalline forms of DNA and various nucleic acid base derivatives. The observed frequencies are discussed with reference to a simple force constant model.     

抗肿瘤药物帕玛度胺类似物的合成

目的：设计并合成抗肿瘤药物帕玛度胺的3位N取代的新型类似物。方法：从3-硝基邻苯二甲酸（2）和N-（叔丁氧羰基）．L-谷氨酰胺（4）出发,经过六步反应得到目标化合物。3-硝基邻苯二甲酸（2）经脱水制得3-硝基邻苯二甲酸酐（3）。用N-（叔丁氧羰基）-L-谷氨酰胺（4）经闭环、脱保护制得3-氨基-2,6-哌啶二酮三氟乙酸盐（6）。4与6经缩合、钯碳催化氢化制得免疫调节剂Pomalidomide（8）,（8）经过酰化得到3-乙酰氨基-N-（2,6-二氧代-3-哌啶基）邻苯二甲酰亚胺（1）。以（4）计总收率约34．9％。结果：得到3-乙酰氨基-N-（2,6-二氧代-3-哌啶基）-邻苯二甲酰亚胺（1）,应用于细胞活性测试。结论：改进了帕玛度胺的合成工艺,得到了3位N乙酰化的新型帕玛度胺类似物（1）,初步研究显示（1）的生物活性与帕玛度胺接近。     

Synthesis of an Alanyl Adenosine Analog

Abstract

The synthetic route towards a 3′-alanyl-3′-amino-3′-deoxyadenosine derivative is described which allows the automated synthesis of 3′-aminoacylated RNA strands.     

Synthesis of a Fluorinated Ganciclovir Analog

Abstract

The synthesis of a new ganciclovir analog with a trifluoromethyl group in the acyclic chain is described.     

Information properties of neuron assemblies

Calculations are made of information parameters of neuron nets with binary plastic synapses of Hebb and Albus which are able to form and produce neuron assemblies. It has been shown that for such neuron nets Hebb synapses are more effective than Albus synapses.