Tsvi Achler
Department of Computer Science
University of Illinois at Urbana Champaign
Thomas M. Seibel Center for Computer Science
201 N. Goodwin Ave
Urbana, IL 61801
Phone: 217 244-7118
Email: achler@uiuc.edu
RESEARCH INTERESTS:
Complex recognition abilities are inherent in even the simplest of organisms, yet is one of the most difficult problems in Computer Science. The responsible neural organization is yet unappreciated. Models based on biological insight can help both understanding of the brain and furthering Computer Science. Thus, I pursue a multidisciplinary approach to recognition-classification with a biological background (Neuroscience, Cognitive Psychology) as a guide and Computer Science as test bed.
One insight involves regulatory feedback connections (where cells feed back to their own inputs). This can be found ubiquitously in the brain including the well-studied olfactory bulb (which has at least two layers of feedback). However, most conventional artificial neural network theories do not incorporate feedback during their 'recognition phase'. Virtually all current network models on some level assume that classification occurs due to connection strengths between cells.
My models focus on regulatory feedback and less on connection strengths. Instead, top-down feedback continuously modifies input activation. The modified input activity is re-distributed to the network and receives feedback on this re-distribution. This is repeated iteratively to determine stimuli relevance.
Because connection weights are not determined per task, the network is more flexible, combinatorially plausible, and learning is simpler. It behaves more informatively in complex scenarios, for example, multiple stimuli present in a scene. It can be used to explain several cognitive phenomena associated with search difficulty, competition and human reaction times. These findings question the traditional approach and offer a more biologically plausible, flexible and a dynamic approach to recognition.
Note: Over the years I have explored several names for this configuration including: Input Shunt Networks, Recurrent Loop Networks, Recurrent Feedback Neural Networks, Input Feedback Networks. However, they all represent the same 'Regulatory Feedback Networks'.REPRESENTATIVE WORK:
Papers / Proceedings
Achler, T., Omar C., Amir, E., Shedding Weights: More With Less, International Joint Conference on Neural Networks (IJCNN), in press. PDF
Achler, T., Amir, E., Input Feedback Networks: Classification and Inference Based on Network Structure, Artificial General Intelligence Proceedings V1: 15-26, 2008. PDF.
Achler, T., Object Classification with Recurrent Feedback Neural Networks, Proc. SPIE, Evolutionary and Bio-inspired Computation: Theory and Applications, Vol. 6563, 2007. PDF
Achler, T., Input Shunt Networks, Neurocomputing, 44-46c: 249-255 Jul 2002.
Talks / Posters
Achler, T., Plasticity Without the Synapse: A Non-Hebbian Account of LTP, Advances in Sensory and Developmental Neuroscience Seminar, UIUC, 2008.
Achler, T., Amir E., Towards a Dynamic Account of Epigenetic Regulation of Expression and Classification, DIMACS Workshop on Computational Neuroscience, Gainesville Florida, 2008.
Achler, T., Object Classification With Recurrent Loop Networks, 10th Tamagawa-Riken Dynamic Brain Forum, Hakuba, Japan, 2007.
Achler T., Input Shunt Networks and Biased Competition, Proceedings of 10th Annual Computational Neuroscience Meeting (CNS), San Francisco and Pacific Grove, California, 2001.
Achler T., Input Shunt Networks, Biased Competition and the Search Task Proceedings of 5th International Conference of Cognitive and Neural Systems (ICCNS), Boston, 2001.