Contact
Giuliano Grossi (Assistant Professor)
Dipartimento di Informatica  Università degli Studi di
Milano
Phone: +39 02503.16262 Curriculum Vitae: CV_en.pdf 
Teaching activity
 current (Italian)

 GPU computing (Laurea Magistrale in Informatica)
 Informazione multimediale (Laurea triennale in Informatica per la comunicazione digitale)
 past

 Metodi per l'elaborazione dei segnali (Laurea Magistrale in Informatica)
 Sistemi e Segnali (Laurea Triennale in Informatica)
 Programmazione in Matlab (Scuola di Fisica Medica)
 Laboratorio di programmazione Java (Laurea Triennale in Informatica)
 Elaborazione numericadei segnali (Laurea Triennale in Informatica)
 Elaborazione di segnali stocastici (Laurea Magistrale in Informatica)
Research interests
 Sparse recovery and regularization methods in signal processing
 LiMapS, kLiMapS: We proposed a fast iterative method for finding sparse solutions to underdetermined linear systems. It is based on a fixedpoint iteration scheme which combines nonconvex Lipschitziantype mappings with canonical orthogonal projectors. The former are aimed at uniformly enhancing the sparseness level by shrinkage effects, the latter are used to project back onto the space of feasible solutions. The iterative process is driven by an increasing sequence of a scalar parameter that mainly contributes to approach the sparsest solutions. It is shown that the minima are locally asymptotically stable for a specific smooth l0norm. Furthermore, it is shown that the points yielded by this iterative strategy are related to the optimal solutions measured in terms of a suitable smooth l1norm. Numerical simulations on phase transition show that the performances of the proposed technique overcome those yielded by well known methods for sparse recovery.
 In the figure we report the 3D phase transitions of LiMapS and other sparsity solvers, such as OMP, CoSaMP, SL0, LASSO and BP. The figure clearly shows the existence of a sharp phase transition that partitions the phase space into a unrecoverable region, with vanishing exactrecovery probability, from an recoverable region in which the probability to recover the optimal sparse vector will eventually approach to one. Qualitatively, it is evident that LiMapS algorithm reaches the best results, having the largest area of high recoverability.
 RSVD: In the sparse representation models, the design of overcomplete dictionaries plays a key role for the effectiveness and applicability in different domains. Recent research has produced several dictionary learning approaches, being proven that dictionaries learnt by data examples significantly outperform structured ones, e.g. wavelet transforms. In this context, learning consists in adapting the dictionary atoms to a set of training signals in order to promote a sparse representation that minimizes the reconstruction error. Finding the best fitting dictionary remains a very difficult task, leaving the question still open. A wellestablished heuristic method for tackling this problem is an iterative alternating scheme, adopted for instance in the wellknown KSVD algorithm. Essentially, it consists in repeating two stages; the former promotes sparse coding of the training set and the latter adapts the dictionary to reduce the error. In this context we presented RSVD, a new method that, while maintaining the alternating scheme, adopts the Orthogonal Procrustes analysis to update the dictionary atoms suitably arranged into groups. Comparative experiments on synthetic data prove the effectiveness of RSVD with respect to well known dictionary learning algorithms such as KSVD, ILSDLA and the online method OSDL. Moreover, experiments on natural data such as ECG compression, EEG sparse representation, and image modeling confirm the RSVD robustness and wide applicability.
 ECG compression: Long duration recordings of ECG signals require high compression ratios, in particular when storing on portable devices. Most of the ECG compression methods in literature are based on wavelet transform while only few of them rely on sparsity promotion models. We proposed a novel ECG signal compression framework based on sparse representation using a set of ECG segments as natural basis. This approach exploits the signal regularity, i.e. the repetition of common patterns, in order to achieve high compression ratio (CR). We applied kLiMapS as fine tuned sparsity solver algorithm guaranteeing the required signal reconstruction quality (PRD). The idea behind the method relies on basis elements drawn from the initial transitory of a signal itself, and the sparsity promotion process applied to its subsequent blocks grabbed by a sliding window. The saved coefficients rescaled in a convenient range, quantized and compressed by a lossless entropybased algorithm. Extensive experiments of our method and of four competitors (namely ARLE, Rajoub, SPIHT, TRE) have been conducted on all the 48 records of MITBIH Arrhythmia Database. Our method achieves average performances that are 3 times higher than the competitor results. In particular the compression ratio gap between our method and the others increases with the PRD growing.
 Face recognition via sparse decomposition
 SSPP problem: Single Sample Per Person (SSPP) Face Recognition is receiving a significant attention due to the challenges it opens especially when conceived for real applications under unconstrained environments. We proposed a solution combining the effectiveness of deep convolutional neural networks (DCNN) feature characterization, the discriminative capability of linear discriminant analysis (LDA), and the efficacy of a sparsity based classifier built on the kLiMapS algorithm. Experiments on the public LFW dataset prove the method robustness to solve the SSPP problem, outperforming several stateoftheart methods.
 FR framework (very few images per subject): For decades, face recognition (FR) has attracted a lot of attention, and several systems have been successfully developed to solve this problem. However, the issue deserves further research e®ort so as to reduce the still existing gap between the computer and human ability in solving it. Among the others, one of the human skills concerns his ability in naturally conferring a degree of reliability" to the face identi ̄cation he carried out. We believe that providing a FR system with this feature would be of great help in real application contexts, making more °exible and treatable the identi ̄cation process. In this spirit, we propose a completely automatic FR system robust to possible adverse illuminations and facial expression variations that provides together with the identity the corresponding degree of reliability. The method promotes sparse coding of multifeature representations with LDA projections for dimensionality reduction, and uses a multistage classi ̄er. The method has been evaluated in the challenging condition of having few (3–5) images per subject in the gallery. Extended experiments on several challenging databases (frontal faces of Extended YaleB, BANCA, FRGC v2.0, and frontal faces of MultiPIE) show that our method outperforms several stateoftheart sparse coding FR systems, thus demon strating its e®ectiveness and generalizability.
 Stochastic models in affective computing
 Deep models for the affective space We draw on a simulationist approach to the analysis of facially displayed emotions  e.g., in the course of a facetoface interaction between an expresser and an observer. At the heart of such perspective lies the enactment of the perceived emotion in the observer. We propose a novel probabilistic framework based on a deep latent representation of a continuous affect space, which can be exploited for both the estimation and the enactment of affective states in a multimodal space (visible facial expressions and physiological signals). The rationale behind the approach lies in the large body of evidence from affective neuroscience showing that when we observe emotional facial expressions, we react with congruent facial mimicry. Further, in more complex situations, affect understanding is likely to rely on a comprehensive representation grounding the reconstruction of the state of the body associated with the displayed emotion. We show that our approach can address such problems in a unified and principled perspective, thus avoiding ad hoc heuristics while minimising learning efforts. Results so far achieved have been assessed by exploiting a publicly available dataset.
 Parallel algorithms on GPUbased architectures

ParCOSNET: Several problems in network biology and
medicine can be cast into a framework where entities are
represented through partially labeled networks, and the aim is
inferring the labels usually binary of the unlabeled part.
Connections represent functional or genetic similarity between
entities, while the labellings often are highly unbalanced, that is
one class is largely underrepresented: for instance in the
automated protein function prediction (AFP) for most Gene Ontology
terms only few proteins are annotated, or in the diseasegene
prioritization problem only few genes are actually known to be
involved in the etiology of a given disease. Imbalanceaware
approaches to accurately predict node labels in biological networks
are thereby required. Furthermore, such methods must be scalable,
since input data can be largesized as, for instance, in the
context of multispecies protein networks.
We proposed a novel semisupervised parallel enhancement of COSNET, an imbalanceaware algorithm build on Hopfield neural model recently suggested to solve the AFP problem. By adopting an efficient representation of the graph and assuming a sparse network topology, we empirically show that it can be efficiently applied to networks with millions of nodes. The key strategy to speed up the computations is to partition nodes into independent sets so as to process each set in parallel by exploiting the power of GPU accelerators. This parallel technique ensures the convergence to asymptotically stable attractors, while preserving the asynchronous dynamics of the original model. Detailed experiments on real data and artificial big instances of the problem highlight scalability and efficiency of the proposed method.
By parallelizing COSNET we achieved on average a speedup of 180x in solving the AFP problem in the S. cerevisiae, Mus musculus and Homo sapiens organisms, while lowering memory requirements. In addition, to show the potential applicability of the method to huge biomolecular networks, we predicted node labels in artificially generated sparse networks involving hundreds of thousands to millions of nodes.  Accelerated VP8 ME: We developed an efficient cooperative interaction between multicore (CPU) and manycore (GPU) resources in the design of a highperformance video encoder. The proposed technique, applied to the wellestabli\shed and highly optimized VP8 encoding format, can achieve a significant speedup with respect to the mostly optimized software encoder (up to $\times$6), with minimum degradation of the visual quality and low processing latency. This result has been obtained through a highly optimized CPUGPU interaction, the exploitation of specific GPU features, and a modified search algorithm specifically adapted to the GPU execution model. Several experimental results are reported and discussed, confirming the effectiveness of the proposed technique. The presented approach, though implemented for the VP8 standard, is of general interest, as it could be applied to any other video encoding scheme.
Projects
 [20192021]
 National project Fondazione Cariplo (N. 20180858). Project title: “Stairway to elders: bridging space, time and emotions in their social environment for wellbeing”, progetto sul bando "Ricerca Sociale sull’invecchiamento: persone, luoghi e relazioni" (Fondazione Cariplo nell’anno 2018).
 [20132016]
 European project FP7ICT201311: Future Networks. Project title: Network Functions asaService over Virtualised Infrastructures (TNOVA), project number 619520.
 [20132016]
 National project Futuro in Ricerca (FIRB) program. Project title: Interpreting emotions: a computational tool integrating facial expressions and biosignals based shape analysis and bayesian networks, Founded by MIUR  Ministero dell'Istruzione dell'Università e della Ricerca.
 [20012002]
 National research project COFIN. Project title: Modelli di calcolo innovativi: metodi sintattici e combinatori, Founded by MIUR  Ministero dell'Istruzione dell'Università e della Ricerca.
 [19982001]
 National project with title Progetto Finalizzato Biotecnologie. Work: Studio e sviluppo di un sistema software per il controllo in tempo reale di esperimenti di misura del calcio intracellulare (Atti del Convegno del Progetto Finalizzato Biotecnologie Genova 2000).
PhD student
 current
 Alessandro D'Amelio
 Alessandro Petrini
 past
 Alessandro Adamo
 Massimo Marchi
Software
 The LiMapS algorithm

A new regularization method for sparse recovery based on a
fixedpoint iteration schema which combines Lipschitziantype
mappings and orthogonal projectors
 LiMaps package for MATLAB
 The kLiMapS algorithm

A new algorithm to solve the sparse approximation problem over
redundant dictionaries based on LiMapS, but retaining the best k
basis (or dictionary) atoms
 kLiMaps package for MATLAB
 The PrunICA algorithm

PrunICA is way to speedup the FastICAlike algorithms by a
controlled random pruning of the input mixtures, both on the entire
mixture or on fixedsize blocks when segmented
 PrunICA package for MATLAB
Papers

 V. Cuculo, A. D'amelio, G. Grossi, R. Lanzarotti, J. Lin
 Robust Single Sample Face Recognition by Sparse Recovery on Learnt Dictionary of DeepCNN Features
 Sensors 19(1), 2019 [doi]

 G. Grossi, P. Paglierani, F. Pedersini, A. Petrini
 Enhanced multicoremanycore interaction in highperformance video encoding
 Journal of RealTime Image Processing, pp. 116, 2018 [doi]

 M. Frasca, G. Grossi, J. Gliozzo, M. Mesiti, M. Notaro, P. Perlasca, A. Petrini, G. Valentini
 A GPUbased algorithm for fast node label learning in large and unbalanced biomolecular networks
 BMC Bioinformatics 19(10), pp. 353, 2018 [doi]

 G. Boccignone, D. Conte, V. Cuculo, A. D’Amelio, G. Grossi, R. Lanzarotti
 Deep Construction of an Affective Latent Space via Multimodal Enactment
 IEEE Transactions on Cognitive and Developmental Systems 10(4), pp. 865880, 2018 [doi]

 M. Bodini, A. D'Amelio, G. Grossi, R. Lanzarotti, J. Lin
 Single Sample Face Recognition by Sparse Recovery of DeepLearned LDA Features
 Advanced Concepts for Intelligent Vision Systems, pp. 297308, 2018 [doi]

 G. Boccignone, M. Bodini, V. Cuculo, G. Grossi
 Predictive sampling of facial expression dynamics driven by a latent action space
 IEEE 14th Int. Conf. on Signal Image Technology Internet Based Systems (SITIS2018), 2018

 G. Boccignone, V. Cuculo, A. D'Amelio, G. Grossi, R. Lanzarotti
 Give Ear to My Face: Modelling Multimodal Attention to Social Interactions
 Computer Vision – ECCV 2018 Workshops, 2018

 A. Adamo, G. Grossi, R. Lanzarotti, J. Lin
 Sparse decomposition by iterating Lipschitziantype mappings
 Theoretical Computer Science 664, pp. 12  28, 2017 [doi]

 G. Grossi, R. Lanzarotti, J. Lin
 Orthogonal Procrustes Analysis for Dictionary Learning in Sparse Linear Representation
 PLOS ONE 12(1), pp. 116, 2017 [doi]

 D. Cusumano, M. Fumagalli, F. Ghielmetti, L. Rossi, G. Grossi, R. Lanzarotti, L. Fariselli, E. De
 Sum signal dosimetry: A new approach for high dose quality assurance with Gafchromic EBT3
 Journal of Applied Clinical Medical Physics, 2017 [doi]

 A. D'Amelio, V. Cuculo, G. Grossi, R. Lanzarotti, J. Lin
 A Note on Modelling a Somatic Motor Space for Affective Facial Expressions
 New Trends in Image Analysis and Processing (ICIAP 2017), pp. 181188, 2017 [doi]

 G. Boccignone, V. Cuculo, G. Grossi, R. Lanzarotti, R. Migliaccio
 Virtual EMG via Facial Video Analysis
 Image Analysis and Processing (ICIAP 2017), pp. 197207, 2017 [doi]

 C. Ceruti, V. Cuculo, A. D'Amelio, G. Grossi, R. Lanzarotti
 Taking the Hidden Route: Deep Mapping of Affect via 3D Neural Networks
 New Trends in Image Analysis and Processing (ICIAP 2017), pp. 189196, 2017 [doi]

 G. Grossi, R. Lanzarotti, J. Lin
 Robust Face Recognition Providing the Identity and Its Reliability Degree Combining Sparse Representation and Multiple Features
 International Journal of Pattern Recognition and Artificial Intelligence 30(10), pp. 1656007, 2016 [doi]

 P. Comi, P. Crosta, M. Beccari, P. Paglierani, G. Grossi, F. Pedersini, A. Petrini
 Hardwareaccelerated highresolution video coding in Virtual Network Functions
 2016 European Conference on Networks and Communications (EuCNC 2016), pp. 3236, 2016 [doi]

 P. Paglierani, G. Grossi, F. Pedersini, A. Petrini
 GPUbased VP8 encoding: Performance in native and virtualized environments
 2016 International Conference on Telecommunications and Multimedia (TEMU 2016), pp. 15, 2016 [doi]

 G. Grossi, R. Lanzarotti, J. Lin
 A Selection Module for LargeScale Face Recognition Systems
 Image Analysis and Processing (ICIAP 2015)  18th International Conference, pp. 529539, 2015 [doi]

 A. Adamo, G. Grossi, R. Lanzarotti, J. Lin
 ECG compression retaining the best natural basis kcoefficients via sparse decomposition
 Biomed. Signal Proc. and Control 15, pp. 1117, 2015 [doi]

 A. Adamo, G. Grossi, R. Lanzarotti, J. Lin
 Robust face recognition using sparse representation in LDA space
 Machine Vision and Applications 26(6), pp. 837847, 2015 [doi]

 G. Grossi, R. Lanzarotti, J. Lin
 Highrate compression of ECG signals by an accuracydriven sparsity model relying on natural basis
 Digital Signal Processing 45, pp. 96106, 2015 [doi]

 A. Adamo, G. Grossi, R. Lanzarotti
 Local features and sparse representation for face recognition with partial occlusions
 IEEE International Conference on Image Processing (ICIP 2013), pp. 30083012, 2013 [doi]

 A. Adamo, G. Grossi, R. Lanzarotti
 Face Recognition in Uncontrolled Conditions Using Sparse Representation and Local Features
 Image Analysis and Processing (ICIAP 2013)  17th International Conference, pp. 3140, 2013 [doi]

 A. Adamo, G. Grossi, R. Lanzarotti
 Sparse Representation Based Classification for Face Recognition by kLiMapS Algorithm
 Image and Signal Processing  5th International Conference (ICISP 2012), pp. 245252, 2012 [doi]

 A. Adamo, G. Grossi
 Sparsity recovery by iterative orthogonal projections of nonlinear mappings
 IEEE Int. Symp. on Signal Processing and Information Technology (ISSPIT 2011), pp. 173178, 2011 [doi]

 A. Adamo, G. Grossi
 A fixedpoint iterative schema for error minimization in ksparse decomposition
 IEEE Int. Symp. on Signal Processing and Information Technology (ISSPIT 2011), pp. 167172, 2011 [doi]

 A. Bertoni, M. Frasca, G. Grossi, G. Valentini
 Learning functional linkage networks with a costsensitive approach
 Proceedings of the 20th Italian Workshop on Neural Nets (WIRN 2010), pp. 5261, 2010 [doi]

 A. Adamo, G. Grossi, F. Pedersini
 Tradeoff between hops and delays in hubbased forwarding in DTNs
 Proceedings of the 3rd IFIP Wireless Days Conference 2010, pp. 15, 2010 [doi]

 A. Adamo, G. Grossi
 Random Pruning of Blockwise Stationary Mixtures for Online BSS
 Latent Variable Analysis and Signal Separation  9th International Conference (LVA/ICA 2010), pp. 213220, 2010 [doi]

 G. Grossi, F. Pedersini
 Hubbetweenness analysis in delay tolerant networks inferred by real traces
 8th International Symposium on Modeling and Optimization in Mobile, AdHoc and Wireless Networks (WiOpt 2010), pp. 318323, 2010

 G. Grossi
 Adaptiveness in Monotone PseudoBoolean Optimization and Stochastic Neural Computation
 Int. J. Neural Syst. 19(4), pp. 241252, 2009 [doi]

 G. Grossi, M. Marchi, E. Pontelli, A. Provetti
 Experimental Analysis of Graphbased Answer Set Computation over Parallel and Distributed Architectures
 J. Log. Comput. 19(4), pp. 697715, 2009 [doi]

 G. Grossi, F. Pedersini
 FPGA implementation of a stochastic neural network for monotonic pseudoBoolean optimization
 Neural Networks 21(6), pp. 872879, 2008 [doi]

 S. Gaito, G. Grossi, F. Pedersini
 A twolevel social mobility model for trace generation
 Proc. of the 9th ACM Int. Symp. on Mobile Ad Hoc Networking and Computing (MobiHoc 2008), pp. 457458, 2008 [doi]

 S. Gaito, G. Grossi, F. Pedersini, P. Rossi
 Experimental validation of a 2level social mobility model in opportunistic networks
 Wireless Days, 2008. WD '08. 1st IFIP, pp. 334338, 2008

 G. Grossi, F. Pedersini
 FPGA Implementation of an Adaptive Stochastic Neural Model
 Artificial Neural Networks  {ICANN} 2007, 17th International Conference, pp. 559568, 2007 [doi]

 S. Gaito, G. Grossi
 Extending Mixture Random Pruning to Nonpolynomial Contrast Functions in FastICA
 Signal Processing and Information Technology (ISSPIT 07), IEEE Int. Symp. on, pp. 334338, 2007 [doi]

 S. Gaito, G. Grossi
 Speeding Up FastICA by Mixture Random Pruning
 Independent Component Analysis and Signal Separation, 7th International Conference (ICA 2007), pp. 185192, 2007 [doi]

 G. Grossi, M. Marchi, E. Pontelli, A. Provetti
 Experiments with answer set computation over parallel and distributed architectures
 4th International Workshop on Answer Set Programming (ASP '07), pp. 720, 2007

 G. Grossi
 A Discrete Adaptive Stochastic Neural Model for Constrained Optimization
 Artificial Neural Networks (ICANN 2006), 16th International Conference, pp. 641650, 2006 [doi]

 G. Grossi, M. Marchi, R. Posenato
 Solving maximum independent set by asynchronous distributed hopfieldtype neural networks
 RAIRO  Theoretical Informatics and Applications 40(2), pp. 371388, 2006 [doi]

 S. Gaito, A. Greppi, G. Grossi
 Random projections for dimensionality reduction in ICA
 International Journal of Applied Science, Engineering and Technology 15, pp. 154158, 2006

 G. Grossi, M. Marchi
 A New Algorithm for Answer Set Computation
 Answer Set Programming, Advances in Theory and Implementation, Proceedings of the 3rd Intl. ASP'05 Workshop, 2005

 G. Grossi, F. Pedersini
 A Stochastic Neural Model for Graph Problems: Software and Hardware Implementation
 Neural Networks and Brain (ICNNB '05). International Conference on, pp. 115120, 2005 [doi]

 A. Bertoni, P. Campadelli, G. Grossi
 A Neural Algorithm for the Maximum Clique Problem: Analysis, Experiments, and Circuit Implementation
 Algorithmica 33(1), pp. 7188, 2002 [doi]

 G. Grossi, R. Posenato
 A Distributed Algorithm for Max Independent Set Problem Based on Hopfield Networks
 Neural Nets, 13th Italian Workshop on Neural Nets (WIRN 2002), pp. 6474, 2002 [doi]

 A. Bertoni, P. Campadelli, G. Grossi
 Solving Min Vertex Cover with Iterated Hopfield Networks
 Neural Nets, 13th Italian Workshop on Neural Nets (WIRN'01), pp. 8795, 2001 [doi]

 A. Bertoni, G. Grossi, A. Provetti, V. Kreinovich, L. Tari
 The Prospect for Answer Sets Computation by a Genetic Model
 Answer Set Programming, Towards Efficient and Scalable Knowledge Representation and Reasoning, Proceedings of the 1st Intl. ASP'01 Workshop, 2001

 A. Bertoni, P. Campadelli, G. Grossi
 An approximation algorithm for the maximum cut problem and its experimental analysis
 Discrete Applied Mathematics 110(1), pp. 312, 2001 [doi]

 A. Bertoni, P. Campadelli, M. Carpentieri, G. Grossi
 A Genetic Model: Analysis and Application to MAXSAT
 Evolutionary Computation 8(3), pp. 291309, 2000 [doi]

 A. Bertoni, P. Campadelli, G. Grossi
 An approximation algorithm for the maximum cut problem and its experimental analysis
 Algorithms and Experiments (ALEX98), pp. 137143, 1998

 M. Alberti, A. Bertoni, P. Campadelli, G. Grossi, R. Posenato
 A Neural Algorithm for MAX2SAT: Performance Analysis and Circuit Implementation
 Neural Networks 10(3), pp. 555560, 1997 [doi]

 A. Bertoni, P. Campadelli, M. Carpentieri, G. Grossi
 Analysis of a Genetic Model
 Proceedings of the 7th International Conference on Genetic Algorithms, pp. 121126, 1997

 G. Grossi
 Sequences of Discrete Hopfield Networks for the Maximum Clique Problem
 Neural Nets, 13th Italian Workshop on Neural Nets (WIRN'97), pp. 139146, 1997 [doi]

 A. Bertoni, P. Campadelli, G. Grossi
 A Discrete Neural Algorithm for the Maximum Clique Problem: Analysis and Circuit Implementation
 Proceedings of the Workshop on Algorithm Engineering (WAE'97), pp. 8491, 1997

 A. Bertoni, P. Campadelli, M. Carpentieri, G. Grossi
 A Genetic Model and the Hopfield Networks
 Artificial Neural Networks (ICANN 96), Int. Conf., pp. 463468, 1996 [doi]

 M. Alberti, A. Bertoni, P. Campadelli, G. Grossi, R. Posenato
 A neural circuit for the maximum 2satisfiability problem
 3rd Euromicro Workshop on Parallel and Distributed Processing (PDP '95), pp. 319323, 1995 [doi]

 M. Alberti, A. Bertoni, P. Campadelli, G. Grossi, R. Posenato
 A neural circuit for the maximum 2satisfiability problem
 Parallel and Distributed Processing. Euromicro Workshop on, pp. 319323, 1995 [doi]