Friedrich T. Sommer
Redwood Neuroscience Institute
1010 El Camino Real, Suite 380
Tel 1-650-8282 Ext. 241
Fax 1-650-8585
fsommer AT rni.org

Friedrich T. Sommer, Ph.D; (Fritz Sommer)

Scientist, Redwood Neuroscience Institute
Faculty Member (Privatdozent) Computer Science Faculty, University of Ulm

Visiting Scholar, University of California Berkeley
cv

 
   page content (links)
- Recent Publications
- Synopsis of Research
- Selected Publications by Theme
- Workshops and Teaching Courses

 



I am interested in the mechanisms underlying human memory and cognition. I usecomputational models of the brain to study this question. An outline of themodeling approach I use and the specific questions I address are described inthe synopsis of research. For a full list of my research publications, see the cv.

Recent Publications

M. Rehn, F.T. Sommer: Early sensory representation in cortex optimizes information contentin sparse sets of neurons.
(2005) submitted

M. Rehn, F. T.Sommer: Rank-based processing of visual input.
(2005) submitted to Neurocomputing

F. T.Sommer, P. Kanerva: Can neural models of cognition benefit from the advantagesof connectionism?

Behavoral BrainSciences (2005) in press

 

F. T. Sommer, T.Wennekers: Synfire chains with conductance-based neurons: internal timing andcoordination with timed input.
Neurocomputing 65-66 (2005)  449 - 454    pdf

D. George, F. T. Sommer: Computing with inter-spike  inverval codes innetworks of integrate and fire neurons.

Neurocomputing 65-66 (2005) 414 - 420    pdf

 

L. M. Martinez, Q.Wang,  R. C. Reid, C. Pillai, J.-M.Alonso, F. T. Sommer, J. A. Hirsch: Receptive field structure varies with layerin the primary visual cortex.
Nature Neuroscience 8 (12) (2005) 372 - 379    pdf

A. Knoblauch, F. T.Sommer: Spike-timing dependent plasticity can form "zero-lag" linksfor cortical oscillations.
Neurocomputing 52-54 (2004) 301 - 306  pdf

M. Rehn, F. T.Sommer: A network for the rapid formation of binary sparse representations ofsensory input.
Technical Report RNI-04-1  (2004)

G. Glatting, F. M.Mottaghy, J. Karitzky, A. Baune, F. T. Sommer, G. B. Landwehrmeyer, S. N.Reske: Improving binding potential analysis in [11C]raclopide PET studies usingcluster analysis.
Medical Physics 31 (4) (2004) 902-906     pdf

J. A. Hirsch, L. M. Martinez, C. Pillai, J.-M. Alonso, Q. Wang, F. T. Sommer:Functionally distinct inhibitory neurons at the first stage of visual corticalprocessing.
Nature Neuroscience 6 (12) (2003) 1300 - 1308    pdf

F. T. Sommer, T. Wennekers: Models of distributed associative memory networksin the brain
Theory in Biosciences (122) (2003) 70 - 86    pdf

Eds: F. T. Sommer, A. Wichert: Exploratory analysis and data modeling infunctional neuroimaging
MIT Press, Boston, MA (2003)    linkto publisher (table of contents, etc.)
 
A. Knoblauch, F. T. Sommer: Synaptic plasticity, conduction delays, andinter-areal phase relations of spike activity in a model of reciprocallyconnected areas
Neurocomputing (52-54) (2003) 301-306    pdf

V. Schmitt, R. Koetter, F. T. Sommer: The impact of thalamo-corticalprojections on activity spread in cortex
Neurocomputing (52-54) (2003) 919-924     pdf


 

Synopsis of Research

Computational models of the brain

   Behavior can be linked to computations in thebrain and studying abstract neural networks can reveal the basic types ofcomputation possible in nerve tissue. However, abstract neural networks areonly crude models of the biophysics of the brain. They can provide a"functional skeleton" in a brain model but more details have to beincluded in order to also quantitatively describe the (bio)physics of neurons, synapses, etc., as observed in Neuro-physiology and -anatomy. There isan fundamental difference between models of physics and computational brainmodels: physics models have usually a single interface to an experimentaldomain, computational brain models have two, they should reflect biophysics andbehavior as well (Sommer et al. 2003).
    The program of neuroscience is to reveal howphysiological/anatomical features of the brain relate to its function.Computational brain models can establish and test hypothesized links betweenbiophysical features and function but in this role they are torn between twoextremes. In order to rule out beforehand as little as possible, experimentalneuroscience explores the full diversity of aspects and a computational brainmodel should reflect the experimental findings as detailed as possible. On theother hand a computational theory of a behavior is basically an algorithm andthe cleanest form to instantiate such a hypothesis is by the simplest neuralnetwork that can do the job efficiently and is not incompatible withneurobiology (following Occam's razor principle).
    Obviously, no single computational brain model can escapethis dilemma. The way I study associative memory function of the brain is toinvestigate a chain of models that vary in the faithfulness of the biophysicaldescription. The starting point of the chain is an abstract neural networkmodel corresponding to the hypothetical function.  Features can be addedto the abstract model step by step, reflecting neurobiological features. Thusthe computational function can be first analyzed in the abstract model. Thepredictions of biophysical brain properties arising from the functionalhypothesis can be assessed in the more detailed models. Qualitative changes inthe model behavior induced by certain model features can be easily traced inthe chain of models. 

Models for local circuits in the brain

   W. James, F. Hayek and D. O. Hebb postulated theories of memory and mentalassociation involving distributed neural representations and synapticplasticity. The most elaborated theory (Hebb, 1949) predicts a concretesynaptic learning rule, the Hebb rule, for learning (distributed) neuralrepresentations of mental entities (thoughts, percepts) called cell assemblies. The directphysiological confirmation of locallearning rules in synapses as proposed in Hebb's theory was a huge success.In other respects, however, the predictive value of Hebb's theory was limitedbecause without a mathematical formulation the predictions were qualitative andcould not be tested in experiments.
    Neuronal associative memories are abstract neural networks that implement the basicmechanisms of learning and association as postulated in Hebb's theory. Neuralassociative memories have been proposed as computational models for localstrongly connected cortical circuits (Palm, 1982, Amit 1989). The computationalfunction is the storage and error-tolerant recall of distributed activitypatterns.  The memory recall is called associative pattern completion ifit involves the completion of a noisy pattern according involving memory.Another recall variant possible in associative memories is pattern recognition(Palm & Sommer 1992) where inputs are just classified as "known"or "unknown".
    A number of different abstract models of associative memoryhas been proposed in the literature. My choice of an abstract model as startingpoint in a chain of computational brain models relies on the assumption thatnature has implemented the hypothetic computational function efficiently.

    Information capacity has become thestandard measure for the efficiency of associative memories. However, thetraditional measures do not take into account all relevant flows of informationduring learning and retrieval. In particular, they neglect the loss due toretrieval errors as well as the information contained in the noisy patternsduring pattern completion tasks. For definitions of information capacity thattake into account all these factors see (Sommer 1993, Palm & Sommer 1996). 

   (Olshausen & Field 1996, Bell & Sejnowski 1996) studied optimal codingstrategies of natural visual scenes. They found that optimal coding of naturalstatistics onto sparse representationsyields neural codes that are in good agreement with the receptive fieldproperties of neurons in primary visual cortex.
    Another argument for sparse memory representations in thebrain follows from the analysis of learning in associative memory. ElisabethGardner's work (1988) revealed a striking match between sparse memoryrepresentations and local learning (Sommer 1993). Basedon her results one can conclude that first, local learning rules store sparsememory patterns more efficiently than nonsparse patterns and second, only forsparse patterns local learning cannot be outperformed by nonlocal learning.Thus, sparse representations of memories naturally arise from optimal localsynaptic learning, a property of synaptic plasticity well confirmed inphysiological studies. Gardner's analysis allows this deep fundamental insight,however, it is not constructive, for instance, it only takes into account thelearning process and not the recall process. Thus, the question remains:

   A general analysis of local learning rules --assessing the capacity of storageand retrieval in a patternassociation task-- is described in (Sommer 1993;Palm & Sommer 1996). For sparse memory patterns, theanalysis characterizes the class of efficient local learning rules. How different superposition schemes for memory traces(in particular, linear superposition as in the Hopfield model and clippedsuperposition as in the Willshaw model) compare in terms of efficiency insparse pattern recognition isanalyzed in (Palm & Sommer 1992; Sommer1993).
    The analyses of sparse associative memory indicate that theclassical Willshaw-Steinbuch model (Steinbuch, 1961; Willshaw et al, 1969) isamong the most efficient models. However, (Palm & Sommer1992; Sommer 1993) also shows for this model thatthe learning provides a higher capacity than the retrieval, i.e., the retrievalin the original model is an information bottleneck. This result raises the questionwhether the Willshaw model can be improved by modified retrieval.
    A modification of the autoassociative Willshaw modelemploying iterative retrieval wasanalyzed in (Sommer 1993; Schwenker,Sommer and Palm 1996). It is shown that the modified retrieval retains theasymptotic information capacity of the original model. However, for (large)finite-sized networks  iterative retrieval has the following advantages:1) A significant increase in recall precision. 2) The asymptotic capacity valuecan be reached in networks of already moderate sizes -- the original model doesnot reach asymptotic performance at practical network sizes. 3) Iterative retrievalis fast. The typical number of required iteration steps is low (<4).
    In bidirectional associative memories (Kosko 1988) withsparse patterns, naive iterative retrieval does not provide the sameimprovement as for autoassociation. (Sommer & Palm 1998,Sommer & Palm 1999) explain why and suggest a novel andvery efficient iterative retrieval in bidirectional associative memories,called crosswise bidirectional retrieval (see also below).

    Having identified efficient instances of sparse associativememory models these can be used in models of neuronal circuits of thebrain. 

    If Hebb's theory were true and brainfunction would be based on cell assemblies, what would their properties be,i.e., how many cells do typically form an assembly and how many assemblies"fit" in a local circuit of cortical tissue?  (Sommer2000)  analyzes a model of a square millimeter of cortex (number ofneurons and connection densities were taken from neuroanatomical studies, cellexcitability was estimated based on physiological studies). The study revealsthat the local synapses are used most efficiently if the size of the assembliesis a few hundred cells and the number of assemblies is in the range between tenand sixty thousand. Due to the incomplete connectivity in the network therearises an interesting extension in functionality: A small set of assemblies(~5) can be recalled simultaneously and not just a single one as in classicalassociative memories.

   Simulation studies with associative networks of conduction-based spikingneurons (two-compartment neurons a la Pinsky & Rinzel, 1994) are describedin (Sommer & Wennekers 2000, Sommer& Wennekers 2001). It is revealed that associative memory recall can becompleted extremely fast, that is, in 25-60ms. Gamma-oscillations can indicateiterative recall (that reaches higher retrieval precision) with latencies of60-260ms.
   

Organization of meso- and macroscopic activity patterns in the brain

    While neural network models described inthe previous section help understanding computations of local brain circuits,cognitive functions ultimately rely on the meso- and macroscopic organizationof neural activity in the brain. The studies in this section address howmacroscopic activity flow can establish cooperative interactions even betweenremote brain regions.

    Reciprocal connectivity is the most commontype of cortico-cortical projections reported by neuroanatomical tracer studies.Thus it is likely that reciprocal connections play an important role inlarge-scale integration of neural representations or cell assemblies. (Sommer & Wennekers 2003) lay out how bidirectionalassociation in reciprocal projections could provide such an integration and howthis ties into earlier work about distributed representations, such as thetheories of Wickelgren, Edelman, Damasio, Mesulam and others.
    Macroscopically distributed cell assemblies would easilyform, if already a single reciprocal connection would express associativememory function. In (Sommer & Wennekers, 2000) abidirectional associative memory model with conductance-based neurons isinvestigated that, in fact, performs efficiently.  A more abstract modelthat is very robust with respect to cross talk --and therefore might be a goodcomputational model of a cortico-cortical projection-- is proposed in (Sommer & Palm 1998, Sommer, Wennekers& Palm 1998, Sommer & Palm1999).

    In recordings of neuronal activity,coherent oscillations mostly occur in phase, even if the recording sites incortex are far apart of each other. For fast (gamma range) oscillationsthis finding is puzzling given the large delay times reported in long-rangeprojections. Modeling studies using reciprocal excitatory couplings with suchdelay times predict anti-phase rather than in-phase correlation. In  (Knoblauch & Sommer 2002, Knoblauch& Sommer 2003) the conditions are studied under which reciprocalcortical connections with realistic delays can express coherent gammaoscillations. It is demonstrated that learning based on spike-timing dependent synapticplasticity (Markram et al. 1997, Poo et al. 1998) can provide robust zero lagcoherence over long-range projections -- zero-lag links.

   Neuronography experiments (MCulloch et al, Pribham et al) revealed thatepileptiform activity elicited by local application of strychnine entailspersistent patterns of activity involving the activity of many brain areas. (Sommer & Koetter 1997, Koetter &Sommer 2000) investigates in a computer model the relation between theanatomy of cortico-cortical  projections and the expression of persistentmacroscopic activity patterns. In the model the connection weights between brainareas can be either simple cortex connectivity schemes such as nearest neighborconnections or data about cortico-cortical projections gathered byneuroanatomical tracer studies and  collatedin the CoCoMac database. Thecomparison between different connectivity schemes shows that neuroanatomicaldata can best explain the measured activity patterns. It is concluded thatlong-range connections are crucial in the formation of patterns that have beenobserved experimentally. Furthermore, the simulations indicate multisynapticreverberating activity propagation and clearly rule out the hypothesis thatjust monosynaptic spread would produce the patterns -- as was speculated in theexperimental literature. (V. Schmitt et al 2003) investigatesthe influence of thalamocortical connections in a similar model.  

    Imaging methods like positron emissiontomography (PET) and functional magnetic resonance (fMRI) provide the first(albeit indirect) windows to macroscopic activation patterns in the workingbrain. The spatio-temporal data sets provided by this methods are usuallysearched for functional activity using regression analysis based on temporalshapes that are estimated based on the timing in the experimental paradigm.However, in short-lasting events and in most cognitive tasks the temporal shapecannot be reliably predicted. In these cases the detection of functionalactivity requires analysis methods based on weaker assumptions about the signalcourse. (Baune et al. 1999) describes a new clusteranalysis method for detecting regions of fMRI activation. The method requiresno information about the time course of the activation and is applied to detecttiming differences in the activation of supplementary motor cortex and motorcortex during a voluntary movement task.
    (Wichert et al. 2003) describes the extension of the methodof Baune et al. for event-related designs. A new method of experimentaldesign/data processing is proposed that yields volumes of data where all slicesare perfectly timed. This avoids the artifacts introduced by usual datapreprocessing methods based on phase-shifting. In (Wichertet al. 2003) the exploratory method is applied to reveal functionalactivity during a n-back working memory task.
    In (Baune et al., 2001, Ruckgaber et al.., 2001) a cluster analysis method was developedto detect microgilia activation which is a very sensitive indicator for brainlesions.

Mathematical analysis of associative memories   

  An attempt to tame the zoo of associative memory modelsproposed in the literature is the Bayesian theory of associative memorydescribed in (Sommer & Dayan 1998). In this theorythe optimal retrieval dynamics can be derived from the uncertainties about theinput pattern and the synaptic weights. Our analysis explains the success ofmany model modifications proposed on heuristic basics, for instance, additionof a ferromagnetic term, of site-dependent thresholds, diagonal terms, variousthreshold strategies, etc.

    The full combinatorial analysis of thefinite Willshaw model can be found in (Sommer & Palm 1999).It predicts distributions of the dendritic potentials and retrieval errors forarbitrary network sizes and all possible types of input noise.

    A general signal-to-noise analysis of locallearning rules is given in (Sommer 1993; Palm & Sommer 1996). The final result is basically oneformula, equation (3.23) in (Palm& Sommer 1996) calculating the S/N for arbitrary learning rules,sparseness levels and input errors. These papers also contain the fullinformation-theoretical treatment of learning and retrieval in associativememories that lead to new definitions of information capacity.

    The asymptotic analysis of the sparseHopfield and Willshaw model is provided in (Palm & Sommer1992). We use elementary analysis information theory and can avoid thecumbersome Replica trick used in the earlier analysis of the Hopfield model(Tsodyks & Feigelman, 1988).


Selected Publications by Theme

(for a completelisting, see cv)

Mechanisms of memory in realistic neural networks

Long-term memory

F. T. Sommer, T. Wennekers : Associative memory in networks ofspiking neurons
Neural Networks 14 (6-7) Special Issue: Spiking Neurons in Neuroscience andTechnology (2001) 825 - 834    pdf
 
F. T. Sommer, T. Wennekers: Modeling studies on thecomputational function of fast temporal structure in cortical circuit activity
Journal of Physiology - Paris 94 (5/6) (2000) 473-488     pdf
    
F. T. Sommer: On cell assemblies in a cortical column
Neurocomputing (32-33) (2000) 517 - 522    pdf

T. Wennekers, F. T. Sommer: Gamma-oscillations supportoptimal retrieval in associative memories of two-compartment neurons
Neurocomputing 26-27 (1999) 573 - 578     pdf

T.Wennekers, F.T.Sommer, G.Palm: IterativeRetrieval in Associative Memories by Threshold Control of Different NeuralModels
In: Supercomputers in Brain Research: From Tomography to Neural Networks WorldScientific Publishing Comp (1995) 301-319    ps

Short-term memory

A. Knoblauch, T. Wennekers, F. T. Sommer: Is voltagedependent synaptic transmission in NMDA receptors a robust mechanism forworking memory?
Neurocomputing (44-46) (2002) 19-24    pdf

U. Vollmer, F. T. Sommer: Coexistence of short and long termmemory in a model network of realistic neurons
Neurocomputing (38-40) (2001) 1031 - 1036    pdf

Physiology and information processing in early vision 

J. A. Hirsch, L. M.Martinez, C. Pillai, J.-M. Alonso, Q. Wang, F. T. Sommer: Functionally distinctinhibitory neurons at the first stage of visual cortical processing.
Nature Neuroscience 6 (12) (2003) 1300 - 1308    pdf

 

Large-scale integration of cortical representations

F. T. Sommer, T.Wennekers : Models of distributed associative memorynetworks in the brain
Theory in Biosciences (122) (2003) 70 - 86    pdf

Associative memory in reciprocal cortico-cortical projections

F. T. Sommer, T. Wennekers : Associative memory in a pairof cortical cell groups with reciprocal projections
Neurocomputing (38-40) (2001) 1575 - 1580    pdf

F. T. Sommer, T. Wennekers, G. Palm: Bidirectional completionof cell assemblies in the cortex
Computational Neuroscience: Trends in Research 1998, Plenum Press, New York,(1998)     ps

Large-scale integration relying on oscillations

A. Knoblauch, F. T. Sommer: Spike-timing dependent plasticitycan form "zero-lag" links for cortical oscillations
submitted to Neurocomputing  (2003)  pdf

A. Knoblauch, F. T. Sommer: Synaptic plasticity, conductiondelays, and inter-areal phase relations of spike activity in a model ofreciprocally connected areas
Neurocomputing (52-54) (2003) 301-306    pdf


 

Large-scale models of cortical activity spread

V. Schmitt, R.Koetter, F. T. Sommer: The impact of thalamo-corticalprojections on activity spread in cortex
Neurocomputing  (2003) (52-54) (2003) 919-924    pdf
 
R. Kötter  and F. T. Sommer: Global relationshipbetween anatomical connectivity and activity propagation in the cerebral cortex
Phil. Trans. R. Soc. Lond. B (355) (2000) 127 - 134    pdf

F.T.Sommer, R. Kötter: Simulating a Network of Cortical AreasUsing Anatomical Connection Data in the Cat
Computational Neuroscience: Trends in Research 1997, Plenum Press, New York(1997) 511-517    ps

R. Koetter, P. Nielsen, J. Dyhrfjeld, F. T. Sommer, G. Northoff: Multi-levelintegration of quantitative neuroanatimical data
Chapter in Computational Neuroanatomy: Principles and Methods.
Ed.: G. A. Ascoli, Humana Press Inc., Totowa, NJ (2002)


 

Theory of sparse associative memory

Bayesian theory of autoassociative memory

F. T. Sommer, P. Dayan: Bayesian Retrieval in AssociativeMemories with Storage Errors
IEEE Transactions on Neural Networks 9 (4) (1998) 705-713    pdf

Bidirectional sparse associative memory

F. T. Sommer, G. Palm: Improved Bidirectional Retrieval ofSparse Patterns Stored by Hebbian Learning
Neural Networks 12 (2) (1999) 281 - 297    pdf

F. T. Sommer, G. Palm: Bidirectional Retrieval fromAssociative Memory
Advances in Neural Information Processing Systems 10, MIT Press, Cambridge, MA(1998) 675 - 681    pdf
    

Analysis of recurrent sparse autoassociative memories

F.Schwenker, F.T.Sommer, G.Palm: Iterative Retrieval ofsparsely coded associative memory patterns
Neural Networks 9 (1996) 445-455    ps 

Analysis of sparse pattern recognition

G.Palm, F.T.Sommer: Information capacity in recurrentMc.Culloch-Pitts networks with sparsely coded memory states
Network 3 (1992) 177-186    pdf
 
G.Palm, F.T.Sommer: Information and pattern capacities in neural associativememories with feedback for sparse memory patterns
In: Neural Network Dynamics, Springer New York (1992). Eds.: J.G.Taylor,E.R.Caianello, R.M.J.Cotterill, J.W.Clark, 3-18
    
Analysis of local learning  rules

G.Palm, F.T.Sommer: Associative data Storage and Retrievalin Neural Nets
In: Models of Neural Networks III, Springer New York (1996) Eds: E.Domany,J.L.van Hemmen, K.Schulten, 79-118    ps

Book, PhD-Thesis (in german)

F. T. Sommer: Theorie neuronaler Assoziativspeicher -Lokales Lernen und iteratives Retrieval von Information
Verlag Hänsel-Hohenhausen (1993) ISBN 3-89349-901-6    ps


Neuroimaging

Edited book

Eds: F. T. Sommer, A. Wichert: Exploratory analysis and datamodeling in functional neuroimaging
MIT Press, Boston, MA (2003)    linkto publisher (table of contents, etc.)
 
General issues of Neuroimaging

F. T. Sommer, J. A. Hirsch, A. Wichert: Theories, dataanalysis and simulation models in neuroimaging - an overview
In  Exploratory analysis and data modeling in functional neuroimaging.
Eds.: F.T. Sommer and A. Wichert,  MIT Press, Boston, MA (2003)    pdf

V. Schmitt, A. Wichert, J. Grothe, F. T. Sommer: The brain positioning software
In: A practical guide of neuroscience databases and associated tools, Ed. R.Koetter, Kluwer, NY (2002)
 
Unsupervised method of detecting functional activity in Neuroimaging

A. Wichert, B. Abler, J. Grothe, H. Walter, F. T. Sommer: Exploratoryanalysis of event-related fMRI demonstrated in a working memory study 
In  Exploratory analysis and data modeling in functional neuroimaging.
Eds.: F.T. Sommer and A. Wichert,  MIT Press, Boston, MA(2003)     pdf

A. Wichert, H. Walter, G. Groen, A. Baune, J. Grothe, A. Wunderlich, F. T.Sommer: Detection of delay selective activity during a working memory task byexplorative data analysis
Neuroimage (13) (2001)  282

A. Baune, F. T. Sommer, M. Erb, D. Wildgruber, B. Kardatzki, G. Palm, W.Grodd:  Dynamical Cluster Analysis of Cortical fMRI Activation
NeuroImage 6 (5) (1999) 477 - 489    pdf

Analysis techniques in Positron Emission Tomography

G. Glatting, F. M. Mottaghy, J. Karitzky, A. Baune, F. T. Sommer, G. B.Landwehrmeyer, S. N. Reske: Improving binding potential analysis in[11C]raclopide PET studies using cluster analysis.
Medical Physics 31 (4) (2004) 902-906     pdf

A. Baune, A. Wichert, G. Glatting, F. T. Sommer: Dynamicalcluster analysis for the detection of microglia activation
in Artificial Neural Nets and Genetic Algorithms. Eds. V. Kurkova, N. C.Stelle, R. Neruda, M. Karny. Springer, Wien (2001) 442 - 445
 
J. Ruckgaber, G. Glatting, J. Karitzky, A. Baune, F. T. Sommer, B. Neumaier, S.N. Reske: Clusteranalyse in derPositronen-Emissions-Tomographie des Hirns mit C-11-PK11195
Nuklearmedizin (40) (2001) A95

Neural associative memories in information technology   

G. Palm, F. Schwenker, F. T. Sommer, A. Strey: Neuralassociative memory
In Associative Processing and Processors, Eds. A. Krikelis and C. C. Weems,IEEE CS Press, Los Alamitos, CA, USA (1997) 307-326    ps
   
F.T.Sommer, F.Schwenker, G.Palm: Assoziative Speicher als Module ininformationsverarbeitenden Systemen
In: Contributions to the Workshop Aspekte Neuronalen Lernens, Eds. L.Cromme, J.Wille, T. Kolb Tech Report, TU Cottbus M-01/1995 (1995)

G.Palm, F.Schwenker, F.T.Sommer: Associative memory and sparse similaritypreserving codes
In: From Statistics to Neural Networks: Theory and Pattern RecognitionApplications, Ed. V.Cherkassky, Springer NATO ASI Series F, New York (1994)282-302

Earlier publications on storage of hydrogenin metal lattices

 

Workshops and TeachingCourses

Recent workshops:

    CNS 2002 workshop: Neural assemblies
    NIPS 2000 workshop: Explorative analysis and data modelingin functional neuroimaging

Graduate teaching (US):

    Theoretical and computational neuroscience(teaching participation at course MCB262/PSYCH290P, UC Berkeley)  

Graduate teaching (semester courses held at University of Ulm):

    Information retrieval and associativememory
    Computational Neuroscience
    Theoretical methods for the interpretation of medicalfunctional imaging data
    Information Retrieval
    Associative memories: conventional and neuronal
    Neural Cell Assemblies


Friedrich T. Sommer -- last update: August, 2003