Poli Interest Group for Machine Learning
Poli Interest Group for Machine Learning (PigML) is an informal meeting (twice a month) of people interested in advanced topics in Machine Learning. We basically read and discuss papers from the ML literature, chosen according the group members interest. Our aim is to discuss ideas and improve all of our understanding.
The scheduled meeting will be pusblished on this page and anyone interested is invited to join the meeting. Topics include but are not limited to:
- Reinforcement Learning
- Evolutionary Computation
- Supervised Learning
- Neural Networks
- Kernel Methods
- Applications of Machine Learning techniques
We strongly encourage anyone to present papers on topics they think can be intereseting for the group. Students are welcome both as audience and as spekers.
Seminar 16/02/2007, h 14:30, Sala Conferenze (DEI)
Paper: Recent Advances in Hierarchical Reinforcement Learning, Special Issue on Reinforcement Learning, Discrete Event Systems journal, vol.13, pp. 41-77, 2003, 
Abstract: Reinforcement learning is bedeviled by the curse of dimensionality: the number of parameters to be learned grows exponentially with the size of any compact encoding of a state. Recent attempts to combat the curse of dimensionality have turned to principled ways of exploiting temporal abstraction, where decisions are not required at each step, but rather invoke the execution of temporally-extended activities which follow their own policies until termination. This leads naturally to hierarchical control architectures and associated learning algorithms. We review several approaches to temporal abstraction and hierarchical organization that machine learning researchers have recently developed. Common to these approaches is a reliance on the theory of semi-Markov decision processes, which we emphasize in our review. We then discuss extensions of these ideas to concurrent activities, multiagent coordination, and hierarchical memory for addressing partial observability. Concluding remarks address open challenges facing the further development of reinforcement learning in a hierarchical setting.
Speaker: Alessandro Lazaric
Seminar 2/03/2007, h 14:30, Sala Seminari (DEI)
Paper: Evolving Neural Networks Through Augmenting Topologies, Evolutionary Computation 10(2):99-127, 2002, 
Abstract: An important question in neuroevolution is how to gain an advantage from evolving neural network topologies along with weights. We present a method, NeuroEvolution of Augmenting Topologies (NEAT), which outperforms the best fixed-topology method on a challenging benchmark reinforcement learning task. We claim that the increased efficiency is due to (1) employing a principled method of crossover of different topologies, (2) protecting structural innovation using speciation, and (3) incrementally growing from minimal structure. We test this claim through a series of ablation studies that demonstrate that each component is necessary to the system as a whole and to each other. What results is signicantly faster learning. NEAT is also an important contribution to GAs because it shows how it is possible for evolution to both optimize and complexify solutions simultaneously, offering the possibility of evolving increasingly complex solutions over generations, and strengthening the analogy with biological evolution.
Speaker: Daniele Loiacono
Seminar 16/03/2007, h 14:30, Sala Seminari (DEI)
Paper: Visualizing the Function Computed by a Feedforward Neural Network, Neural Computation, Vol 12, Issue 6, Pages: 1337 - 1353, 2000, 
Abstract: A method for visualizing the function computed by a feedforward neural network is presented. It is most suitable for models with continuous inputs and a small number of outputs, where the output function is reasonably smooth, as in regression and probabilistic classification tasks. The visualization makes readily apparent the effects of each input and the way in which the functions deviate from a linear function. The visualization can also assist in identifying interactions in the fitted model. The method uses only the input-output relationship and thus can be applied to any predictive statistical model, including bagged and committee models, which are otherwise difficult to interpret. The visualization method is demonstrated on a neural network model of how the risk of lung cancer is affected by smoking and drinking.
Speaker: Rossella Blatt
Seminar 28/03/2007, h 14:30, Sala Seminari (DEI)
Paper: A tutorial on hidden Markov models and selected applications inspeech recognition, Rabiner L., Proceedings of the IEEE, Vol 77, Pages: 257-286, 1989 
Abstract: Hidden Markov Models have been introduced and studied in the late 1960's but they have become increasingly popular in the last several years. An HMM describe an undergoing Markov porocess that can be observed from another output Markov process. The main reason for witch they are so important is that they can model a given stocastic process in an unsupervised fashion. This paper present basic theory to learn such kind of models and many applications to speech recognition.
Speaker: Simone Tognetti
Seminar 18/04/2007, h 14:30, Sala Conferenze (DEI)
Paper: TEMPLAR: a wavelet-based framework for pattern learning and analysis, Scott, C.; Nowak, R.D., Signal Processing, IEEE Transactions on Acoustics, Speech, and Signal Processing, Vol. 52, 2004 
Recovering a pattern or image from a collection of noisy and misaligned observations is a challenging problem that arises in image processing and pattern recognition. This paper presents an automatic, wavelet-based approach to this problem. Despite the success of wavelet decompositions in other areas of statistical signal and image processing, most wavelet-based image models are inadequate for modeling patterns in images, due to the presence of unknown transformations (e.g., translation, rotation, location of lighting source) inherent in pattern observations. Our framework takes advantage of the efficient image representations afforded by wavelets while accounting for unknown translations and rotations. In order to learn the parameters of our model from training data, we introduce Template Learning from Atomic Representations (TEMPLAR): a novel template learning algorithm. The problem solved by TEMPLAR is the recovery of a pattern template from a collection of noisy, randomly translated, and rotated observations of the pattern. TEMPLAR employs minimum description length (MDL) complexity regularization to learn a template with a sparse representation in the wavelet domain. We discuss several applications, including template learning, pattern classification, and image registration.
Speaker: Giacomo Boracchi
Seminar 06/06/2007, h 15:30, Sala Seminari (DEI)
Paper: The Dynamics of AdaBoost: Cyclic Behavior and Convergence of Margins, Cynthia Rudin, Ingrid Daubechies, Robert E. Schapire, The Journal of Machine Learning Research, Vol. 5, 2004 
Abstract: This talk will present AdaBoost, an algorithm for improving the performance of classifiers by aggregating them; some properties of AdaBoost will be discussed, particularly its dynamics, both in terms of loss minimization and margin maximization, and the possible arising of cycles.
Speaker: Bernardo Dal Seno
Seminar 04/07/2007, h 14:30, Sala Seminari (DEI)
Seminar: Fuzzy k-NN Lung Cancer Identification by an Electronic Nose
Abstract: We present a method to recognize the presence of lung cancer in individuals by classifying the olfactory signal acquired through an electronic nose based on an array of MOS sensors. We analyzed the breath of 101 persons, of which 58 as control and 43 suffering from different types of lung cancer (primary and not) at different stages. In order to find the components able to discriminate between the two classes ‘healthy’ and ‘sick’ as best as possible and to reduce the dimensionality of the problem, we extracted the most significative features and projected them into a lower dimensional space, using Nonparametric Linear Discriminant Analysis. Finally, we used these features as input to a pattern classification algorithm, based on Fuzzy k-Nearest Neighbors (Fuzzy k-NN). The observed results, all validated using cross-validation, have been satisfactory achieving an accuracy of 92.6%, a sensitivity of 95.3% and a specificity of 90.5%. These results put the electronic nose as a valid implementation of lung cancer diagnostic technique, being able to obtain excellent results with a non invasive, small, low cost and very fast instrument.
Speaker: Rossella Blatt