BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/Los_Angeles BEGIN:DAYLIGHT DTSTART:20250309T030000 TZOFFSETFROM:-0800 TZOFFSETTO:-0700 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:PDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20251102T010000 TZOFFSETFROM:-0700 TZOFFSETTO:-0800 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:PST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20250715T190656Z UID:B90D0D41-B635-4637-8737-0B3BE1470587 DTSTART;TZID=America/Los_Angeles:20250716T110000 DTEND;TZID=America/Los_Angeles:20250716T120000 DESCRIPTION:Optimization problems subject to hard constraints are common in time-critical applications such as autonomous driving\, communications\, networking\, and power grid operation. However\, existing iterative solver s often face difficulties in solving these problems in real-time. In this talk\, we advocate a machine learning approach -- to employ NN's approxima tion capability to learn the input-solution mapping of a problem and then pass new input through the NN to obtain a quality solution\, orders of mag nitude faster than iterative solvers. To date\, the approach has achieved promising empirical performance and exciting theoretical development. A fu ndamental issue\, however\, is to ensure NN solution feasibility with resp ect to the hard constraints\, which is non-trivial due to inherent NN pred iction errors. To this end\, we present two approaches\, predict-and-recon struct and homeomorphic projection\, to ensure NN solution strictly satisf ies the equality and inequality constraints\, respectively. In particular\ , homeomorphic projection is a low-complexity scheme to guarantee NN solut ion feasibility for optimization over any set homeomorphic to a unit ball\ , covering all compact convex sets and certain classes of nonconvex sets. The idea is to (i) learn a minimum distortion homeomorphic mapping between the constraint set and a unit ball using an invertible NN (INN)\, and the n (ii) perform a simple bisection operation concerning the unit ball so th at the INN-mapped final solution is feasible with respect to the constrain t set with minor distortion-induced optimality loss. We prove the feasibil ity guarantee and bound the optimality loss under mild conditions. Simulat ion results\, including those for computation-heavy SDP problems and non-c onvex AC-OPF problems for grid operations\, show that homeomorphic project ion outperforms existing methods in solution feasibility and run-time comp lexity\, while achieving similar optimality loss. We will also discuss ope n problems and future directions.\n\nSpeaker(s): \, Minghua\n\nRoom: 3038\ , Bldg: Macleod Building\, 2356 Main Mall\, Vancouver\, British Columbia\, Canada\, V6T1Z4 LOCATION:Room: 3038\, Bldg: Macleod Building\, 2356 Main Mall\, Vancouver\, British Columbia\, Canada\, V6T1Z4 ORGANIZER:lelewang@ece.ubc.ca SEQUENCE:10 SUMMARY:Machine Learning for Real-Time Constrained Optimization URL;VALUE=URI:https://events.vtools.ieee.org/m/493042 X-ALT-DESC:Description: <br /><p>Optimization problems subject to hard cons traints are common in time-critical applications such as autonomous drivin g\, communications\, networking\, and power grid operation. However\, exis ting iterative solvers often face difficulties in solving these problems i n real-time. In this talk\, we advocate a machine learning approach -- to employ NN's approximation capability to learn the input-solution mapping o f a problem and then pass new input through the NN to obtain a quality sol ution\, orders of magnitude faster than iterative solvers. To date\, the a pproach has achieved promising empirical performance and exciting theoreti cal development. A fundamental issue\, however\, is to ensure NN solution feasibility with respect to the hard constraints\, which is non-trivial du e to inherent NN prediction errors. To this end\, we present two approache s\, predict-and-reconstruct and homeomorphic projection\, to ensure NN sol ution strictly satisfies the equality and inequality constraints\, respect ively. In particular\, homeomorphic projection is a low-complexity scheme to guarantee NN solution feasibility for optimization over any set homeomo rphic to a unit ball\, covering all compact convex sets and certain classe s of nonconvex sets. The idea is to (i) learn a minimum distortion homeomo rphic mapping between the constraint set and a unit ball using an invertib le NN (INN)\, and then (ii) perform a simple bisection operation concernin g the unit ball so that the INN-mapped final solution is feasible with res pect to the constraint set with minor distortion-induced optimality loss. We prove the feasibility guarantee and bound the optimality loss under mil d conditions. Simulation results\, including those for computation-heavy S DP problems and non-convex AC-OPF problems for grid operations\, show that homeomorphic projection outperforms existing methods in solution feasibil ity and run-time complexity\, while achieving similar optimality loss. We will also discuss open problems and future directions.</p> END:VEVENT END:VCALENDAR