{"id":235,"date":"2013-05-20T00:21:21","date_gmt":"2013-05-20T00:21:21","guid":{"rendered":"http:\/\/web.eecs.umich.edu\/~girasole\/wordpress\/?page_id=235"},"modified":"2026-02-27T16:46:34","modified_gmt":"2026-02-27T16:46:34","slug":"students","status":"publish","type":"page","link":"https:\/\/web.eecs.umich.edu\/~girasole\/?page_id=235","title":{"rendered":"SPADA lab"},"content":{"rendered":"

Signal Processing Algorithm Design and Analysis<\/strong><\/h3>\n

\"SPADAOur lab studies algorithms for statistical signal processing and machine learning with applications in data analysis, computer vision, environmental monitoring, image processing, control systems, power grids, genetic expression data analysis, consumer preference modeling, and computer network analysis. We are interested in algorithmic design using principles from optimization theory, as well as mathematical analysis answering questions regarding algorithmic convergence behavior and performance, required sample complexity, algorithmic robustness. See the projects page<\/a> for descriptions of some of our research areas and the publications page<\/a> for our research papers.<\/p>\n

You can find Professor Balzano’s mentoring plan here.<\/p>\n

For prospective postdocs or students at all levels interested in joining SPADA lab, please read to\u00a0the end of this page.<\/p>\n

SPADA lab April 2024:
\n<\/b><\/p>\n

\"SPADA<\/h3>\n

Ph.D. Students<\/strong><\/h3>\n

I have the pleasure of working with the following outstanding students and postdocs. Listed is their most recent publication with the SPADA lab.
\n
\nWhat is the most effective advertising for a lawyer? Search engine promotion of the website of a lawyer, lawyer or law firm is the most effective way to promote advertising services. The bulk of potential customers are looking for information through search engines. Therefore, special attention law firm marketing<\/a> should be paid to SEO optimization and website maintenance. The higher the site rises in the top, the more often people will find it, and the more potential customers will start to go to its pages, asking for help and advice.<\/span><\/p>\n

Alex Ritchie | <\/strong>website<\/a> | (co-advised with Clay Scott<\/a>)<\/em><\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\tRitchie<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tapa<\/span>\n\t\t1<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 235<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%22TZMM3WBZ%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Ritchie%20et%20al.%22%2C%22parsedDate%22%3A%222022-08-16%22%2C%22numChildren%22%3A2%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BRitchie%2C%20A.%2C%20Balzano%2C%20L.%2C%20Kessler%2C%20D.%2C%20Sripada%2C%20C.%20S.%2C%20%26amp%3B%20Scott%2C%20C.%20%282022%29.%20%26lt%3Bi%26gt%3BSupervised%20PCA%3A%20A%20Multiobjective%20Approach%26lt%3B%5C%2Fi%26gt%3B%20%28arXiv%3A2011.05309%29.%20arXiv.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2011.05309%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2011.05309%26lt%3B%5C%2Fa%26gt%3B%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Supervised%20PCA%3A%20A%20Multiobjective%20Approach%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alexander%22%2C%22lastName%22%3A%22Ritchie%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%22%2C%22lastName%22%3A%22Balzano%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Daniel%22%2C%22lastName%22%3A%22Kessler%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Chandra%20S.%22%2C%22lastName%22%3A%22Sripada%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Clayton%22%2C%22lastName%22%3A%22Scott%22%7D%5D%2C%22abstractNote%22%3A%22Methods%20for%20supervised%20principal%20component%20analysis%20%28SPCA%29%20aim%20to%20incorporate%20label%20information%20into%20principal%20component%20analysis%20%28PCA%29%2C%20so%20that%20the%20extracted%20features%20are%20more%20useful%20for%20a%20prediction%20task%20of%20interest.%20Prior%20work%20on%20SPCA%20has%20focused%20primarily%20on%20optimizing%20prediction%20error%2C%20and%20has%20neglected%20the%20value%20of%20maximizing%20variance%20explained%20by%20the%20extracted%20features.%20We%20propose%20a%20new%20method%20for%20SPCA%20that%20addresses%20both%20of%20these%20objectives%20jointly%2C%20and%20demonstrate%20empirically%20that%20our%20approach%20dominates%20existing%20approaches%2C%20i.e.%2C%20outperforms%20them%20with%20respect%20to%20both%20prediction%20error%20and%20variation%20explained.%20Our%20approach%20accommodates%20arbitrary%20supervised%20learning%20losses%20and%2C%20through%20a%20statistical%20reformulation%2C%20provides%20a%20novel%20low-rank%20extension%20of%20generalized%20linear%20models.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2011.05309%22%2C%22date%22%3A%222022-08-16%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2011.05309%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2011.05309%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22J2WEX3C9%22%5D%2C%22dateModified%22%3A%222022-09-14T23%3A31%3A20Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Ritchie, A., Balzano, L., Kessler, D., Sripada, C. S., & Scott, C. (2022). Supervised PCA: A Multiobjective Approach<\/i> (arXiv:2011.05309). arXiv. https:\/\/doi.org\/10.48550\/arXiv.2011.05309<\/a><\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n

Rachel Newton<\/strong> | website<\/a> | (co-advised with Peter Seiler<\/a>)
\n<\/em><\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\tNewton<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tapa<\/span>\n\t\t1<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 235<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%22D3LM67BL%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Newton%20et%20al.%22%2C%22parsedDate%22%3A%222023-12-19%22%2C%22numChildren%22%3A2%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BNewton%2C%20R.%2C%20Du%2C%20Z.%2C%20Seiler%2C%20P.%2C%20%26amp%3B%20Balzano%2C%20L.%20%282023%29.%20Optimality%20of%20POD%20for%20Data-Driven%20LQR%20With%20Low-Rank%20Structures.%20%26lt%3Bi%26gt%3BIEEE%20Control%20Systems%20Letters%26lt%3B%5C%2Fi%26gt%3B%2C%20%26lt%3Bi%26gt%3B8%26lt%3B%5C%2Fi%26gt%3B%2C%2085%26%23x2013%3B90.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FLCSYS.2023.3344147%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.1109%5C%2FLCSYS.2023.3344147%26lt%3B%5C%2Fa%26gt%3B%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22journalArticle%22%2C%22title%22%3A%22Optimality%20of%20POD%20for%20Data-Driven%20LQR%20With%20Low-Rank%20Structures%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Rachel%22%2C%22lastName%22%3A%22Newton%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhe%22%2C%22lastName%22%3A%22Du%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Peter%22%2C%22lastName%22%3A%22Seiler%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%22%2C%22lastName%22%3A%22Balzano%22%7D%5D%2C%22abstractNote%22%3A%22The%20optimal%20state-feedback%20gain%20for%20the%20Linear%20Quadratic%20Regulator%20%28LQR%29%20problem%20is%20computationally%20costly%20to%20compute%20for%20high-order%20systems.%20Reduced-order%20models%20%28ROMs%29%20can%20be%20used%20to%20compute%20feedback%20gains%20with%20reduced%20computational%20cost.%20However%2C%20the%20performance%20of%20this%20common%20practice%20is%20not%20fully%20understood.%20This%20letter%20studies%20this%20practice%20in%20the%20context%20of%20data-driven%20LQR%20problems.%20We%20show%20that%2C%20for%20a%20class%20of%20LQR%20problems%20with%20low-rank%20structures%2C%20the%20controllers%20designed%20via%20their%20ROM%2C%20based%20on%20the%20Proper%20Orthogonal%20Decomposition%20%28POD%29%2C%20are%20indeed%20optimal.%20Experimental%20results%20not%20only%20validate%20our%20theory%20but%20also%20demonstrate%20that%20even%20with%20moderate%20perturbations%20on%20the%20low-rank%20structure%2C%20the%20incurred%20suboptimality%20is%20mild.%22%2C%22date%22%3A%2212%5C%2F19%5C%2F2023%22%2C%22section%22%3A%22%22%2C%22partNumber%22%3A%22%22%2C%22partTitle%22%3A%22%22%2C%22DOI%22%3A%2210.1109%5C%2FLCSYS.2023.3344147%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22https%3A%5C%2F%5C%2Fieeexplore.ieee.org%5C%2Fabstract%5C%2Fdocument%5C%2F10365496%22%2C%22PMID%22%3A%22%22%2C%22PMCID%22%3A%22%22%2C%22ISSN%22%3A%222475-1456%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22DZFDBB6V%22%5D%2C%22dateModified%22%3A%222024-12-12T22%3A11%3A12Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Newton, R., Du, Z., Seiler, P., & Balzano, L. (2023). Optimality of POD for Data-Driven LQR With Low-Rank Structures. IEEE Control Systems Letters<\/i>, 8<\/i>, 85\u201390. https:\/\/doi.org\/10.1109\/LCSYS.2023.3344147<\/a><\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n

Can Yaras<\/strong> | website<\/a> | (co-advised with Qing Qu<\/a>)<\/em><\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\tYaras<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tapa<\/span>\n\t\t1<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 235<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%223A9Y78SV%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Xu%20et%20al.%22%2C%22parsedDate%22%3A%222026-02-05%22%2C%22numChildren%22%3A3%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BXu%2C%20A.%20S.%2C%20Yaras%2C%20C.%2C%20Asato%2C%20M.%2C%20Qu%2C%20Q.%2C%20%26amp%3B%20Balzano%2C%20L.%20%282026%29.%20%26lt%3Bi%26gt%3BEmergent%20Low-Rank%20Training%20Dynamics%20in%20MLPs%20with%20Smooth%20Activations%26lt%3B%5C%2Fi%26gt%3B%20%28arXiv%3A2602.06208%29.%20arXiv.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2602.06208%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2602.06208%26lt%3B%5C%2Fa%26gt%3B%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22Emergent%20Low-Rank%20Training%20Dynamics%20in%20MLPs%20with%20Smooth%20Activations%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Alec%20S.%22%2C%22lastName%22%3A%22Xu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Can%22%2C%22lastName%22%3A%22Yaras%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Matthew%22%2C%22lastName%22%3A%22Asato%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qing%22%2C%22lastName%22%3A%22Qu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%22%2C%22lastName%22%3A%22Balzano%22%7D%5D%2C%22abstractNote%22%3A%22Recent%20empirical%20evidence%20has%20demonstrated%20that%20the%20training%20dynamics%20of%20large-scale%20deep%20neural%20networks%20occur%20within%20low-dimensional%20subspaces.%20While%20this%20has%20inspired%20new%20research%20into%20low-rank%20training%2C%20compression%2C%20and%20adaptation%2C%20theoretical%20justification%20for%20these%20dynamics%20in%20nonlinear%20networks%20remains%20limited.%20%25compared%20to%20deep%20linear%20settings.%20To%20address%20this%20gap%2C%20this%20paper%20analyzes%20the%20learning%20dynamics%20of%20multi-layer%20perceptrons%20%28MLPs%29%20under%20gradient%20descent%20%28GD%29.%20We%20demonstrate%20that%20the%20weight%20dynamics%20concentrate%20within%20invariant%20low-dimensional%20subspaces%20throughout%20training.%20Theoretically%2C%20we%20precisely%20characterize%20these%20invariant%20subspaces%20for%20two-layer%20networks%20with%20smooth%20nonlinear%20activations%2C%20providing%20insight%20into%20their%20emergence.%20Experimentally%2C%20we%20validate%20that%20this%20phenomenon%20extends%20beyond%20our%20theoretical%20assumptions.%20Leveraging%20these%20insights%2C%20we%20empirically%20show%20there%20exists%20a%20low-rank%20MLP%20parameterization%20that%2C%20when%20initialized%20within%20the%20appropriate%20subspaces%2C%20matches%20the%20classification%20performance%20of%20fully-parameterized%20counterparts%20on%20a%20variety%20of%20classification%20tasks.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2602.06208%22%2C%22date%22%3A%222026-02-05%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2602.06208%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2602.06208%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22J2WEX3C9%22%5D%2C%22dateModified%22%3A%222026-02-16T21%3A58%3A43Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Xu, A. S., Yaras, C., Asato, M., Qu, Q., & Balzano, L. (2026). Emergent Low-Rank Training Dynamics in MLPs with Smooth Activations<\/i> (arXiv:2602.06208). arXiv. https:\/\/doi.org\/10.48550\/arXiv.2602.06208<\/a><\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n

Javier Salazar Cavazos<\/strong> | website<\/a> | (co-advised with Jeff Fessler<\/a>)<\/em><\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\tCavazos<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tapa<\/span>\n\t\t2<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 235<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%22QIB5RKK9%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Cavazos%20et%20al.%22%2C%22parsedDate%22%3A%222025-06-01%22%2C%22numChildren%22%3A2%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BCavazos%2C%20J.%20S.%2C%20Fessler%2C%20J.%20A.%2C%20%26amp%3B%20Balzano%2C%20L.%20%282025%29.%20%26lt%3Bi%26gt%3BALPCAHUS%3A%20Subspace%20Clustering%20for%20Heteroscedastic%20Data%26lt%3B%5C%2Fi%26gt%3B%20%28arXiv%3A2505.18918%29.%20arXiv.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2505.18918%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2505.18918%26lt%3B%5C%2Fa%26gt%3B%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22ALPCAHUS%3A%20Subspace%20Clustering%20for%20Heteroscedastic%20Data%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Javier%20Salazar%22%2C%22lastName%22%3A%22Cavazos%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Jeffrey%20A.%22%2C%22lastName%22%3A%22Fessler%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%22%2C%22lastName%22%3A%22Balzano%22%7D%5D%2C%22abstractNote%22%3A%22Principal%20component%20analysis%20%28PCA%29%20is%20a%20key%20tool%20in%20the%20field%20of%20data%20dimensionality%20reduction.%20Various%20methods%20have%20been%20proposed%20to%20extend%20PCA%20to%20the%20union%20of%20subspace%20%28UoS%29%20setting%20for%20clustering%20data%20that%20come%20from%20multiple%20subspaces%20like%20K-Subspaces%20%28KSS%29.%20However%2C%20some%20applications%20involve%20heterogeneous%20data%20that%20vary%20in%20quality%20due%20to%20noise%20characteristics%20associated%20with%20each%20data%20sample.%20Heteroscedastic%20methods%20aim%20to%20deal%20with%20such%20mixed%20data%20quality.%20This%20paper%20develops%20a%20heteroscedastic-focused%20subspace%20clustering%20method%2C%20named%20ALPCAHUS%2C%20that%20can%20estimate%20the%20sample-wise%20noise%20variances%20and%20use%20this%20information%20to%20improve%20the%20estimate%20of%20the%20subspace%20bases%20associated%20with%20the%20low-rank%20structure%20of%20the%20data.%20This%20clustering%20algorithm%20builds%20on%20K-Subspaces%20%28KSS%29%20principles%20by%20extending%20the%20recently%20proposed%20heteroscedastic%20PCA%20method%2C%20named%20LR-ALPCAH%2C%20for%20clusters%20with%20heteroscedastic%20noise%20in%20the%20UoS%20setting.%20Simulations%20and%20real-data%20experiments%20show%20the%20effectiveness%20of%20accounting%20for%20data%20heteroscedasticity%20compared%20to%20existing%20clustering%20algorithms.%20Code%20available%20at%20https%3A%5C%2F%5C%2Fgithub.com%5C%2Fjaviersc1%5C%2FALPCAHUS.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22arXiv%22%2C%22archiveID%22%3A%22arXiv%3A2505.18918%22%2C%22date%22%3A%222025-06-01%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2505.18918%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2505.18918%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22J2WEX3C9%22%5D%2C%22dateModified%22%3A%222025-06-17T12%3A09%3A45Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Cavazos, J. S., Fessler, J. A., & Balzano, L. (2025). ALPCAHUS: Subspace Clustering for Heteroscedastic Data<\/i> (arXiv:2505.18918). arXiv. https:\/\/doi.org\/10.48550\/arXiv.2505.18918<\/a><\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n

Soo Min Kwon<\/strong> | website<\/a> | (co-advised with Qing Qu<\/a>)<\/em><\/p>\n

\n\n\t\t1399621<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\tKwon<\/span>\n\t\t<\/span>\n <\/span>\n\t\t1<\/span>\n\t\tapa<\/span>\n\t\t1<\/span>\n\t\tdate<\/span>\n\t\tdesc<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n\t\t<\/span>\n <\/span>\n 235<\/span>\n\t\thttps:\/\/web.eecs.umich.edu\/~girasole\/wp-content\/plugins\/zotpress\/<\/span>\n\n\t\t
\n\t\t\t
\n\t\t\t\t%7B%22status%22%3A%22success%22%2C%22updateneeded%22%3Afalse%2C%22instance%22%3Afalse%2C%22meta%22%3A%7B%22request_last%22%3A0%2C%22request_next%22%3A0%2C%22used_cache%22%3Atrue%7D%2C%22data%22%3A%5B%7B%22key%22%3A%22FZGX4TR7%22%2C%22library%22%3A%7B%22id%22%3A1399621%7D%2C%22meta%22%3A%7B%22creatorSummary%22%3A%22Balzano%20et%20al.%22%2C%22parsedDate%22%3A%222026%22%2C%22numChildren%22%3A2%7D%2C%22bib%22%3A%22%26lt%3Bdiv%20class%3D%26quot%3Bcsl-bib-body%26quot%3B%20style%3D%26quot%3Bline-height%3A%202%3B%20padding-left%3A%201em%3B%20text-indent%3A-1em%3B%26quot%3B%26gt%3B%5Cn%20%20%26lt%3Bdiv%20class%3D%26quot%3Bcsl-entry%26quot%3B%26gt%3BBalzano%2C%20L.%2C%20Ding%2C%20T.%2C%20Haeffele%2C%20B.%20D.%2C%20Kwon%2C%20S.%20M.%2C%20Qu%2C%20Q.%2C%20Wang%2C%20P.%2C%20Wang%2C%20Z.%2C%20%26amp%3B%20Yaras%2C%20C.%20%282026%29.%20%26lt%3Bi%26gt%3BAn%20Overview%20of%20Low-Rank%20Structures%20in%20the%20Training%20and%20Adaptation%20of%20Large%20Models%26lt%3B%5C%2Fi%26gt%3B%20%28arXiv%3A2503.19859%29.%20Accepted%20to%20Signal%20Processing%20Magazine%2C%20special%20issue%20on%20the%20Mathematics%20of%20Deep%20Learning.%20%26lt%3Ba%20class%3D%26%23039%3Bzp-DOIURL%26%23039%3B%20href%3D%26%23039%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2503.19859%26%23039%3B%26gt%3Bhttps%3A%5C%2F%5C%2Fdoi.org%5C%2F10.48550%5C%2FarXiv.2503.19859%26lt%3B%5C%2Fa%26gt%3B%26lt%3B%5C%2Fdiv%26gt%3B%5Cn%26lt%3B%5C%2Fdiv%26gt%3B%22%2C%22data%22%3A%7B%22itemType%22%3A%22preprint%22%2C%22title%22%3A%22An%20Overview%20of%20Low-Rank%20Structures%20in%20the%20Training%20and%20Adaptation%20of%20Large%20Models%22%2C%22creators%22%3A%5B%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Laura%22%2C%22lastName%22%3A%22Balzano%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Tianjiao%22%2C%22lastName%22%3A%22Ding%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Benjamin%20D.%22%2C%22lastName%22%3A%22Haeffele%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Soo%20Min%22%2C%22lastName%22%3A%22Kwon%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Qing%22%2C%22lastName%22%3A%22Qu%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Peng%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Zhangyang%22%2C%22lastName%22%3A%22Wang%22%7D%2C%7B%22creatorType%22%3A%22author%22%2C%22firstName%22%3A%22Can%22%2C%22lastName%22%3A%22Yaras%22%7D%5D%2C%22abstractNote%22%3A%22The%20rise%20of%20deep%20learning%20has%20revolutionized%20data%20processing%20and%20prediction%20in%20signal%20processing%20and%20machine%20learning%2C%20yet%20the%20substantial%20computational%20demands%20of%20training%20and%20deploying%20modern%20large-scale%20deep%20models%20present%20significant%20challenges%2C%20including%20high%20computational%20costs%20and%20energy%20consumption.%20Recent%20research%20has%20uncovered%20a%20widespread%20phenomenon%20in%20deep%20networks%3A%20the%20emergence%20of%20low-rank%20structures%20in%20weight%20matrices%20and%20learned%20representations%20during%20training.%20These%20implicit%20low-dimensional%20patterns%20provide%20valuable%20insights%20for%20improving%20the%20efficiency%20of%20training%20and%20fine-tuning%20large-scale%20models.%20Practical%20techniques%20inspired%20by%20this%20phenomenon%2C%20such%20as%20low-rank%20adaptation%20%28LoRA%29%20and%20training%2C%20enable%20significant%20reductions%20in%20computational%20cost%20while%20preserving%20model%20performance.%20In%20this%20paper%2C%20we%20present%20a%20comprehensive%20review%20of%20recent%20advances%20in%20exploiting%20low-rank%20structures%20for%20deep%20learning%20and%20shed%20light%20on%20their%20mathematical%20foundations.%20Mathematically%2C%20we%20present%20two%20complementary%20perspectives%20on%20understanding%20the%20low-rankness%20in%20deep%20networks%3A%20%28i%29%20the%20emergence%20of%20low-rank%20structures%20throughout%20the%20whole%20optimization%20dynamics%20of%20gradient%20and%20%28ii%29%20the%20implicit%20regularization%20effects%20that%20induce%20such%20low-rank%20structures%20at%20convergence.%20From%20a%20practical%20standpoint%2C%20studying%20the%20low-rank%20learning%20dynamics%20of%20gradient%20descent%20offers%20a%20mathematical%20foundation%20for%20understanding%20the%20effectiveness%20of%20LoRA%20in%20fine-tuning%20large-scale%20models%20and%20inspires%20parameter-efficient%20low-rank%20training%20strategies.%20Furthermore%2C%20the%20implicit%20low-rank%20regularization%20effect%20helps%20explain%20the%20success%20of%20various%20masked%20training%20approaches%20in%20deep%20neural%20networks%2C%20ranging%20from%20dropout%20to%20masked%20self-supervised%20learning.%22%2C%22genre%22%3A%22%22%2C%22repository%22%3A%22Accepted%20to%20Signal%20Processing%20Magazine%2C%20special%20issue%20on%20the%20Mathematics%20of%20Deep%20Learning%22%2C%22archiveID%22%3A%22arXiv%3A2503.19859%22%2C%22date%22%3A%222026%22%2C%22DOI%22%3A%2210.48550%5C%2FarXiv.2503.19859%22%2C%22citationKey%22%3A%22%22%2C%22url%22%3A%22http%3A%5C%2F%5C%2Farxiv.org%5C%2Fabs%5C%2F2503.19859%22%2C%22language%22%3A%22%22%2C%22collections%22%3A%5B%22DZFDBB6V%22%5D%2C%22dateModified%22%3A%222026-02-17T00%3A25%3A20Z%22%7D%7D%5D%7D<\/span>\n\n\t\t\t\t
\n
\n
Balzano, L., Ding, T., Haeffele, B. D., Kwon, S. M., Qu, Q., Wang, P., Wang, Z., & Yaras, C. (2026). An Overview of Low-Rank Structures in the Training and Adaptation of Large Models<\/i> (arXiv:2503.19859). Accepted to Signal Processing Magazine, special issue on the Mathematics of Deep Learning. https:\/\/doi.org\/10.48550\/arXiv.2503.19859<\/a><\/div>\n<\/div>\n\t\t\t\t<\/div>\n\t\t\t<\/div>\n\t\t<\/div>\n\t<\/div>\n\n\n

Elvin Tseng<\/strong> | website<\/a><\/p>\n

Laya Pullela<\/strong> | website<\/a> | (co-advised with Minji Kim)<\/em><\/p>\n

Jessica Jiang<\/strong> | website<\/a> | (co-advised with Clay Scott)<\/em><\/p>\n

Gavin Kunesh<\/strong> | website<\/a> | (co-advised with Al Hero)<\/em><\/p>\n

Postdocs<\/strong><\/h3>\n

(None currently)<\/p>\n

Master’s and Undergraduate Students<\/strong><\/h3>\n

Matt Asato
\nLinglong Meng<\/strong><\/p>\n

<\/p>\n

Former Ph.D. Students and Postdocs<\/strong><\/h3>\n

Peng Wang<\/strong> | website<\/a> | (co-advised with Qing Qu<\/em>)
\n<\/em>Next position: Assistant Professor at University of Macau<\/p>\n

Zhe Du<\/strong> | website<\/a> | (co-advised with Necmiye Ozay<\/a>)
\n<\/em>Defended November 2022, “Learning, Control, and Reduction for Markov Jump Systems”
\nNext position: Postdoc at UC Riverside.<\/p>\n

Kyle Gilman<\/strong> | website<\/a>
\nDefended October 2022, “Scalable Algorithms Using Optimization on Orthogonal Matrix Manifolds”
\nNext positions: Applied AI\/ML Senior Associate at Chase Bank. Member of Technical Staff at MIT Lincoln Laboratory.<\/p>\n

Davoud Ataee Tarzanagh <\/strong>| website<\/a>
\n<\/strong>Next positions: Postdoctoral scholar at University of Pennsylvania. AI Scientist at Samsung.<\/p>\n

Haroon Raja <\/strong>| website<\/a>
\n<\/strong>Next positions: Postdoctoral scholar at Tufts. AI Research Scientist at Eli Lilly.
\n<\/strong><\/p>\n

Amanda Bower <\/strong>| website<\/a>
\nDefended October 2020, “Dealing with Intransitivity, Non-Convexity, and Algorithmic Bias in Preference Learning”
\nNext position: Twitter’s ML Ethics, Transparency, and Accountability (META) group.<\/p>\n

Ali Soltani-Tehrani <\/strong>| website<\/a>
\n<\/strong>Next position: Associate Principal Data Scientist at AstraZeneca.<\/p>\n

Dejiao Zhang<\/strong> | website<\/a>
\nDefended May 2019, “Extracting Compact Knowledge from Massive Data”
\nNext positions: Applied research scientist at Amazon Web Services. Senior staff research scientist at Figma.<\/p>\n

David Hong <\/strong>| website<\/a> | (co-advised with Jeff Fessler<\/a>)<\/em>
\nDefended March 2019, “Learning Low-Dimensional Models for Heterogeneous Data”
\nNext positions: Postdoctoral scholar at Penn – Wharton Statistics Department. Assistant Professor at the University of Delaware.<\/p>\n

Greg Ongie<\/strong> | website<\/a>
\nNext positions: Postdoctoral scholar at University of Chicago – Statistics and Computer Science Departments. Assistant professor of Mathematics at Marquette University.<\/p>\n

John Lipor<\/strong> | website <\/a>
\nDefended September 2017, “Sensing Structured Signals with Active and Ensemble Methods”
\nNext positions: Assistant and then Associate Professor in the Portland State University ECE Department<\/p>\n

Former MS lab members:
\n<\/strong><\/h4>\n

Yutong Wang\u00a0
\n<\/em>Rishhabh Naik
\nNisarg Trivedi
\nGeoffrey Fortman
\nPengyu Xiao
\nSaket Dewangan
\nJenna King<\/p>\n

Former undergraduate researchers:
\n<\/strong><\/h4>\n

Ian Steele
\nJake Hume
\nIman Malik
\nAustin Xu
\nEli Smith
\nAndrew Gitlin
\nBob Malinas
\nNora Farouk
\nWilliam Zhang
\nRichard Ortman<\/p>\n

Prospective students and postdocs:<\/span><\/strong><\/p>\n

If you are interested in joining my research lab, either working on a small project, a Ph.D. thesis, or a postdoctoral project, please read this information.<\/b><\/p>\n

In the SPADA lab, we enjoy working with students and collaborators who have an enthusiastic curiosity for mathematics and algorithms and their applications in machine learning and signal processing. Our work will draw on tools from probability, linear algebra, and functional analysis as well as contemporary mathematics. You must also demonstrate independent thinking, patience, and integrity. For students already at Michigan, both graduate and undergraduate: Please email me with your CV and the types of projects you’d be interested in, and we can set up a time to meet during my office hours. For prospective postdocs: Please email me with your CV and a research statement, and highlight our shared interests and potential collaborative projects. For prospective graduate students: Please start by applying to the ECE Michigan graduate program<\/a>. Include my name in your research statement along with reasons why you’d be interested in working with me. I am not actively looking for new students in the fall of 2022, however, I am always open to working with truly outstanding students.<\/p>\n

Please make your email subject line “Joining the Balzano lab” to show me you have read this. I will do my best to respond as soon as I can. Since I am busy taking care of my current students, it may take several weeks before I find the time, so I encourage your patience. You will appreciate these priorities if you end up at Michigan. If I have not replied in a few months time, you can assume your background was not a good fit for the lab. Best of luck to you in your search for a research mentor.<\/p>\n

SPADA lab September 2022:
\n<\/b><\/p>\n

\"SPADA<\/p>\n

SPADA lab January 2021:<\/b><\/p>\n

<\/p>\n

SPADA lab April 2019:<\/b><\/p>\n

<\/p>\n

SPADA lab December 2017:<\/b><\/p>\n

A huge industry has grown up producing generic drugs which contain the same active ingredients as the branded drugs. One does not get an erection just by taking Kamagra xn--brablpiller-18a.com\/kamagra-oral-jelly-online<\/a>. During this period external sexual stimulus would cause the penis to become erect.<\/p>\n

<\/p>\n

The abstract is considered an insignificant student work in comparison with coursework or diploma works. Therefore, we pay less attention to writing it and often do not care about the originality of the text and the harmony of the composition. Nevertheless, writing an essay is an integral part of the educational process. Working on 500 word essay<\/a>, we first of all develop skills for creating more serious work. That is why you need to be able to write an essay with high quality.<\/p>\n

What is an abstract?<\/p>\n

Before considering the stages of writing a competent, solid abstract, let’s find out what is hidden behind this word.<\/p>\n

In Russland wurden jedoch die Einfuhr des Impfstoffs von Pfizer und seine Verwendung bisher von Roszdravnadzor verboten. Das Verbot gilt f\u00fcr alle Impfstoffe starker-mann.com<\/a> zur Pr\u00e4vention von Coronavirus-Infektionen, die nicht staatlich zugelassen sind. Wie oft verschreiben \u00c4rzte nicht zugelassene Arzneimittel?<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Signal Processing Algorithm Design and Analysis Our lab studies algorithms for statistical signal processing and machine learning with applications in data analysis, computer vision, environmental monitoring, image processing, control systems, power grids, genetic expression data analysis, consumer preference modeling, and computer network analysis. We are interested in algorithmic design using principles from optimization theory, as […]<\/p>\n","protected":false},"author":2,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"open","ping_status":"closed","template":"","meta":[],"_links":{"self":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/235"}],"collection":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=235"}],"version-history":[{"count":100,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/235\/revisions"}],"predecessor-version":[{"id":1028,"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=\/wp\/v2\/pages\/235\/revisions\/1028"}],"wp:attachment":[{"href":"https:\/\/web.eecs.umich.edu\/~girasole\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=235"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}