BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.2.3.1//EN TZID:Europe/Paris X-WR-TIMEZONE:Europe/Paris BEGIN:VEVENT UID:8544@i2m.univ-amu.fr DTSTART;TZID=Europe/Paris:20250310T140000 DTEND;TZID=Europe/Paris:20250310T150000 DTSTAMP:20250307T100714Z URL:https://www.i2m.univ-amu.fr/evenements/seminaire-de-thomas-schatz-2/ SUMMARY:Thomas Schatz (AMU): Séminaire de Thomas Schatz DESCRIPTION:Thomas Schatz: Title\n\nUnbiased Estimation from Unbalanced Hie rarchical Samples\, with Applications to the Evaluation of Representation Learning Algorithms\n\nAbstract\n\nClassical U-statistics' theory generali ze statistical estimation theory from averages of functions of one i.i.d. sample (for example the Euclidean norm is a function of one sample\, if sa mples are real vectors) to averages of function of several i.i.d. samples (for example\, the Euclidean distance is a function of two samples if samp les are real vectors). It notably provides Uniformly Minimum Variance Unbi ased Estimators (UMVUEs) for many statistical models of practical relevanc e. But what happens if we are not working with simple i.i.d. samples?\n\nI n this talk\, motivated by applications to representational geometry analy sis in machine learning and cognitive (neuro)science [1-6]\, we will consi der the case of possibly heavily unbalanced hierarchical samples. We will start with the non-trivial issue of defining hierarchical samples with eno ugh generality to cover the application cases of interest. We will then se e that i.i.d. samples can be represented as graphs with the special proper ty that all graph isomorphisms are also graph automorphisms and that the s implicity of the classical theory depends on this property. In the case of unbalanced hierarchical samples\, this property no longer holds and we wi ll discuss what are the proper ways to generalize and what can be salvaged of the classical theory in this case. We will conclude with some computat ional considerations.\n\nReferences\n\n[1] Aarre Laakso and Garrison Cottr ell. “Content and cluster analysis: assessing representational similarit y in neural systems”. Philosophical psychology 13.1 (2000)\, pp. 47–76 .\n\n[2] James V Haxby\, M Ida Gobbini\, Maura L Furey\, Alumit Ishai\, Je nnifer L Schouten\, and Pietro Pietrini. “Distributed and overlapping re presentations of faces and objects in ventral temporal cortex”. Science 293.5539 (2001)\, pp. 2425–2430.\n\n[3] Nikolaus Kriegeskorte\, Marieke Mur\, and Peter A Bandettini. “Representational similarity analysis-conn ecting the branches of systems neuroscience”. Frontiers in systems neuro science 2 (2008)\, p. 249.\n[4] Thomas Schatz\, Vijayaditya Peddinti\, Fra ncis Bach\, Aren Jansen\, Hynek Hermansky and Emmanuel Dupoux. “Evaluati ng speech features with the minimal-pair ABX task: Analysis of the classic al MFC/PLP pipeline”. Proceeding of INTERSPEECH 2013.\n \;\n\n[5] Co rinna Cortes\, Mehryar Mohri\, and Afshin Rostamizadeh. “Algorithms for learning kernels based on centered alignment”. The Journal of Machine Le arning Research 13 (2012)\, pp. 795–828.\n\n[6] Simon Kornblith\, Mohamm ad Norouzi\, Honglak Lee\, and Geoffrey Hinton. “Similarity of Neural Ne twork Representations Revisited”. Proceedings of ICML 2019. CATEGORIES:Séminaire,Statistique LOCATION:I2M Saint-Charles - Salle séminaire\, Aix-Marseille Université I nstitut de Mathématiques de Marseille (I2M) - UMR 7373 \, 3 place Victor Hugo Case 19\, Marseille Cedex 3\, 13331\, France X-APPLE-STRUCTURED-LOCATION;VALUE=URI;X-ADDRESS=Aix-Marseille Université I nstitut de Mathématiques de Marseille (I2M) - UMR 7373 \, 3 place Victor Hugo Case 19\, Marseille Cedex 3\, 13331\, France;X-APPLE-RADIUS=100;X-TIT LE=I2M Saint-Charles - Salle séminaire:geo:0,0 END:VEVENT BEGIN:VTIMEZONE TZID:Europe/Paris X-LIC-LOCATION:Europe/Paris BEGIN:STANDARD DTSTART:20241027T020000 TZOFFSETFROM:+0200 TZOFFSETTO:+0100 TZNAME:CET END:STANDARD END:VTIMEZONE END:VCALENDAR