From dilsat.yuksel at strath.ac.uk Fri Mar 7 14:06:19 2025 From: dilsat.yuksel at strath.ac.uk (Dilsat Yuksel) Date: Fri, 7 Mar 2025 14:06:19 +0000 Subject: [msp-interest] [MSP101] Zanzi Mihejevs: A canonical bidirectional typing discipline through polarised System L (LT412, 3pm Mon 10-Mar) Message-ID: Hi, On Monday, the MSP101 talk will be given by Zanzi Mihejevs (GlAiVe Research). See the details of the talk below. Best, Dilsat ********************************************************************** Date: Monday, 10 March at 15:00 Room: LT412, Livingstone Tower Zoom link: https://strath.zoom.us/j/85449272187?pwd=0qNkQqybiaKgBpbEBRv38x11x4db5n.1 Live stream on Youtube: https://www.youtube.com/@msp101strathclyde4 Speaker: Zanzi Mihejevs Title: A canonical bidirectional typing discipline through polarised System L Abstract: What is the relationship between polarity and bidirectional typing? It has long been observed that there is a connection between the two, but the precise relationship has remained unclear. Moreover, it has been argued that the link itself is a red herring, and that bidirectional typing is better explained not by polarity but by chirality - the duality between producers and consumers. In this talk we will look at Polarised System L, a type theory that combines both dualities - the positive fragment is driven by a cut between a primitive producer and a pattern, and the negative fragment is driven by a cut between a primitive consumer and a co-pattern. Remarkably, linear System L admits a canonical bidirectional typing discipline based on a combination of ideas from both standard and co-contextual typing, giving us a bi-contextual typing algorithm. We will see how this lets us equip a type system based on full classical linear logic - containing all four connectives and derivable implication and co-implication - with a bidirectional discipline where all typing annotations are exclusively limited to shifts between sythesisable and checkable expressions. ********************************************************************** -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.strath.ac.uk/archives/msp-interest/attachments/20250307/0fdac882/attachment.html From dilsat.yuksel at strath.ac.uk Thu Mar 13 16:27:22 2025 From: dilsat.yuksel at strath.ac.uk (Dilsat Yuksel) Date: Thu, 13 Mar 2025 16:27:22 +0000 Subject: [msp-interest] [MSP101] Ross Horne: Additives without Weakening (LT412, 3pm Mon 17-Mar) Message-ID: Hi, On Monday, the MSP101 talk will be given by Ross Horne (MSP, StrathCyber). See the details of the talk below. Best, Dilsat ********************************************************************** Date: Monday, 17 March at 15:00 Room: LT412, Livingstone Tower Zoom link: https://strath.zoom.us/j/85449272187?pwd=0qNkQqybiaKgBpbEBRv38x11x4db5n.1 Live stream on Youtube: https://www.youtube.com/@msp101strathclyde4 Speaker: Ross Horne Title: Additives without Weakening Abstract: As we know, in linear logic, conjunction and disjunction decompose into separate multiplicative and additive forms. Only additives are idempotent in general (A + A --o A). The additives also have some properties such as weakening (A & B --o A). By dropping weakening, the additives further decompose into infinitely many sub-additive operators. I explain how I spotted these sub-additive operators when studying nominal quantifiers that can similarly be decomposed into new nominal quantifiers. I found these new additives curious since the resulting operators have properties that are sound with respect to probability distributions, effectively internalising probability distributions in logic. I end by giving a taste of the proof theory of the sub-additives. A novelty is that, due to how sub-additives control certain distributivity properties that play an essential role in established cut elimination techniques, a new proof technique must be invented. ********************************************************************** -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.strath.ac.uk/archives/msp-interest/attachments/20250313/f38bcc2a/attachment.html From fredrik.nordvall-forsberg at strath.ac.uk Wed Mar 19 09:32:55 2025 From: fredrik.nordvall-forsberg at strath.ac.uk (Fredrik Nordvall Forsberg) Date: Wed, 19 Mar 2025 09:32:55 +0000 Subject: [msp-interest] =?utf-8?q?=5BMSP101=5D_Petar_Veli=C4=8Dkovi=C4=87=3A_Generalisat?= =?utf-8?q?ion_in_LLMs_=28LT210=2C_4pm_TODAY_19/3=29?= Message-ID: Dear all, Petar Veli?kovi?, one of Neil's visitors from Google DeepMind this week, has kindly offered to give us a talk at 4pm TODAY Wednesday. This will be a whiteboard talk in LT210, but there will of course also be an online option (see below). Hope to see you there, Fred ********************************* Topic: Generalisation in LLMs Speaker: Petar Veli?kovi? (Google DeepMind) Room: LT210, Livingstone Tower Zoom link: https://strath.zoom.us/j/85449272187?pwd=0qNkQqybiaKgBpbEBRv38x11x4db5n.1 ********************************* Neil says: > Hi All > > Apologies for short notice but Petar Velickovic is happy to give an > informal whiteboard talk Weds 4pm > > Topic = A general talk about our recent works on generalisation in llms > > Location = LT210 > > cheers > neil -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.strath.ac.uk/archives/msp-interest/attachments/20250319/949cd805/attachment.html