BEGIN:VCALENDAR VERSION:2.0 PRODID:-//128.220.36.25//NONSGML kigkonsult.se iCalcreator 2.26.9// CALSCALE:GREGORIAN METHOD:PUBLISH X-FROM-URL:https://www.clsp.jhu.edu X-WR-TIMEZONE:America/New_York BEGIN:VTIMEZONE TZID:America/New_York X-LIC-LOCATION:America/New_York BEGIN:STANDARD DTSTART:20231105T020000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RDATE:20241103T020000 TZNAME:EST END:STANDARD BEGIN:DAYLIGHT DTSTART:20240310T020000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RDATE:20250309T020000 TZNAME:EDT END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT UID:ai1ec-20117@www.clsp.jhu.edu DTSTAMP:20240329T083021Z CATEGORIES;LANGUAGE=en-US:Seminars CONTACT: DESCRIPTION:
Abstract
\nNeural sequence generation systems oftentimes generate sequences by searching for the most likely se quence under the learnt probability distribution. This assumes that the mo st likely sequence\, i.e. the mode\, under such a model must also be the b est sequence it has to offer (often in a given context\, e.g. conditioned on a source sentence in translation). Recent findings in neural machine tr anslation (NMT) show that the true most likely sequence oftentimes is empt y under many state-of-the-art NMT models. This follows a large list of oth er pathologies and biases observed in NMT and other sequence generation mo dels: a length bias\, larger beams degrading performance\, exposure bias\, and many more. Many of these works blame the probabilistic formulation of NMT or maximum likelihood estimation. We provide a different view on this : it is mode-seeking search\, e.g. beam search\, that introduces many of t hese pathologies and biases\, and such a decision rule is not suitable for the type of distributions learnt by NMT systems. We show that NMT models spread probability mass over many translations\, and that the most likely translation oftentimes is a rare event. We further show that translation d istributions do capture important aspects of translation well in expectati on. Therefore\, we advocate for decision rules that take into account the entire probability distribution and not just its mode. We provide one exam ple of such a decision rule\, and show that this is a fruitful research di rection.
\nBiography
\nI am an assistant professor (UD) in natural language processing at the Institute for Logic\, Language and Computation where I lead the Probabilistic Language L earning group.
\nMy work concerns the design of models and algor ithms that learn to represent\, understand\, and generate language data. E xamples of specific problems I am interested in include language modelling \, machine translation\, syntactic parsing\, textual entailment\, text cla ssification\, and question answering.
\nI also develop techniques to approach general machine learning problems such as probabilistic inferenc e\, gradient and density estimation.
\nMy interests sit at the inter section of disciplines such as statistics\, machine learning\, approximate inference\, global optimization\, formal languages\, and computational li nguistics.
\n\n
DTSTART;TZID=America/New_York:20210419T120000 DTEND;TZID=America/New_York:20210419T131500 LOCATION:via Zoom SEQUENCE:0 SUMMARY:Wilker Aziz (University of Amsterdam) “The Inadequacy of the Mode in Neural Machine Translation” URL:https://www.clsp.jhu.edu/events/wilker-aziz-university-of-amsterdam/ X-COST-TYPE:free X-TAGS;LANGUAGE=en-US:2021\,April\,Aziz END:VEVENT BEGIN:VEVENT UID:ai1ec-23304@www.clsp.jhu.edu DTSTAMP:20240329T083021Z CATEGORIES;LANGUAGE=en-US:Seminars CONTACT: DESCRIPTION:
Abstract
\nTransformers are essential to pretraining. As we approach 5 years of BERT\, the connection between a ttention as architecture and transfer learning remains key to this central thread in NLP. Other architectures such as CNNs and RNNs have been used t o replicate pretraining results\, but these either fail to reach the same accuracy or require supplemental attention layers. This work revisits the semanal BERT result and considers pretraining without attention. We consid er replacing self-attention layers with recently developed approach for lo ng-range sequence modeling and transformer architecture variants. Specific ally\, inspired by recent papers like the structured space space sequence model (S4)\, we use simple routing layers based on state-space models (SSM ) and a bidirectional model architecture based on multiplicative gating. W e discuss the results of the proposed Bidirectional Gated SSM (BiGS) and p resent a range of analysis into its properties. Results show that architec ture does seem to have a notable impact on downstream performance and a di fferent inductive bias that is worth exploring further.
\nBi ography
\n