Search
Search Funnelback University
- Refined by:
- Date: Past fortnight
1 -
38 of
38
search results for katalk:za31 24 / / / / / |u:mlg.eng.cam.ac.uk
where 0
match all words and 38
match some words.
Results that match 1 of 2 words
-
Understanding the Bethe Approximation: When and How can it go Wrong?
https://mlg.eng.cam.ac.uk/adrian/pabc.pdf19 Jun 2024: 100. Maximum coupling strength y. BethelocalBethecycleBethemargTRWlocalTRWcycleTRWmarg. log partition error. 2 8 16 24 320. ... 2 8 16 24 320. 0.1. 0.2. 0.3. 0.4. Maximum coupling strength y. -
Blind Justice: Fairness with Encrypted Sensitive Attributes
https://mlg.eng.cam.ac.uk/adrian/poster_ICML18_BlindJustice.pdf19 Jun 2024: Bank. Datasets and Feasibility. Adult Bank COMPAS German SQFn training examples 214 215 212 29 216d features 51 62 7 24 23p sensitive attributes 1 1 7 1 1certification 802 ms -
Clamping Variables and Approximate Inference
https://mlg.eng.cam.ac.uk/adrian/newsclamp.pdf19 Jun 2024: 0.1. 0.2. 0.3. 0.4. max interaction strength W 24 / 19. -
Revisiting the Limits of MAP Inference by MWSS on Perfect Graphs
https://mlg.eng.cam.ac.uk/adrian/slides-revisit.pdf19 Jun 2024: Revisiting the Limits of MAP Inference by MWSSon Perfect Graphs. Adrian WellerUniversity of Cambridge. CP 2015Cork, Ireland. Slides and full paper athttp://mlg.eng.cam.ac.uk/adrian/. 1 / 21. Motivation: undirected graphical models (MRFs). • -
Clamping Variables and Approximate Inference
https://mlg.eng.cam.ac.uk/adrian/slides_msr2.pdf19 Jun 2024: 24 / 21. Error in log Z vs number of clamps: complete graphs. ... 24 / 21. Time (secs) vs error in log Z for various methods. -
Directed and Undirected Graphical Models
https://mlg.eng.cam.ac.uk/adrian/2018-MLSALT4-AW1-models.pdf19 Jun 2024: 23 / 26. Directed and undirected models are different. 24 / 26. ... Directed and undirected models are different. With <3 edges,. 24 / 26. -
Junction Tree, BP and Variational Methods
https://mlg.eng.cam.ac.uk/adrian/2018-MLSALT4-AW3-approx.pdf19 Jun 2024: 24 / 32. The log-partition function log Z. Since D(q‖p) 0, we have. ... a finite convex hull 3. M(G) = conv {φ(e), e X m} (5.24). -
The Case for Process Fairness in Learning:Feature Selection for ...
https://mlg.eng.cam.ac.uk/adrian/grgic.pdf19 Jun 2024: Age 46% 56% 32%Gender 24% 50% 22%Race 24% 35% 15%. Table 3: Comparing judgment of fairness of each feature, of 54 male users, when they have differentknowledge about the ... Age 42% 63% 33%Gender 24% 54% 23%Race 18% 43% 18%. Table 5: Comparing judgment -
Bounding the Integrality Distance ofLP Relaxations for Structured…
https://mlg.eng.cam.ac.uk/adrian/OPT2016_paper_3.pdf19 Jun 2024: 24)Using the same approach to upper-bound Equation 22, we assume, without loss of generality, that(δ(µI ) θ) µ̃L (δ(µI ) θ. ′) ... m. (25). Using Equation 25 to upper-bound 22, and Equation 24 to upper-bound 23, we have that. -
Structured Prediction Models for Chord Transcription of Music Audio…
https://mlg.eng.cam.ac.uk/adrian/icmla09adrian.pdf19 Jun 2024: To its right, 0-1 adds the features of the previous frame, so in this model each frame is represented by 24 dimensions. -
ML-IRL: Machine Learning in Real Life Workshop at ICLR ...
https://mlg.eng.cam.ac.uk/adrian/ML_IRL_2020-CLUE.pdf19 Jun 2024: 6). LSATAl. (7). COMPASEp. (6). COMPASAl. (5) Total (24). Prolific Unc. -
Transparency: Motivations and Challenges? Adrian…
https://mlg.eng.cam.ac.uk/adrian/transparency.pdf19 Jun 2024: Notions of fairness beyond equality, and the role ofrandomness in fairness, were recently explored [24, 74]. ... Advances in experimental social psychology 24, 319–359 (1991). 26. Hammond, R., Axelrod, R.: The evolution of ethnocentrism. -
Understanding the Bethe Approximation: When and How can it ...
https://mlg.eng.cam.ac.uk/adrian/abc.pdf19 Jun 2024: FA8750-14-C-0005. 2 8 16 24 320. 20. 40. 60. 80. 100. ... Maximum coupling strength y. BethelocalBethecycleBethemargTRWlocalTRWcycleTRWmarg. (a) log partition error. 2 8 16 24 320. -
Now You See Me (CME): Concept-based Model Extraction
https://mlg.eng.cam.ac.uk/adrian/AIMLAI20-CME.pdf19 Jun 2024: Model ExtractionModel extraction techniques use rules [20, 21, 22], de-cision trees [23, 24], or other more readily explainablemodels [25] to approximate complex models, in orderto study their behaviour. ... 24] M. Sato, H. Tsukimoto, Rule extraction -
Ode to an ODE Krzysztof Choromanski ∗Robotics at Google ...
https://mlg.eng.cam.ac.uk/adrian/NeurIPS20-ODEtoODE.pdf19 Jun 2024: MIT Press, 2016. [24] Aditya Grover, Eric Wang, Aaron Zweig, and Stefano Ermon. ... In Proceedings of the 33nd International Conference on Machine Learning,ICML 2016, New York City, NY, USA, June 19-24, 2016, pages 2034–2042, 2016. -
Evaluating and Aggregating Feature-based Model Explanations
https://mlg.eng.cam.ac.uk/adrian/IJCAI20_EvaluatingAndAggregating.pdf19 Jun 2024: µC(f, gSHAP) 1.94 0.26 1.36 0.36 2.33 0.23µC(f, gAVA) 1.93 0.24 1.24 0.32 2.61 0.29. -
An Introduction to LP Relaxations for MAP Inference
https://mlg.eng.cam.ac.uk/adrian/2018-MLSALT4-AW2-LP.pdf19 Jun 2024: x3. x4. x5. x6. s. balanced almost balanced(attractive up to flipping) 24 / 41. -
Gauged Mini-Bucket Elimination for Approximate Inference Sungsoo Ahn…
https://mlg.eng.cam.ac.uk/adrian/Gauge_for_Holder_Inference.pdf19 Jun 2024: We call these methods collectively Z-invariant methods. See [23, 24, 25] for discussions ofthe differences and relations between these methods. ... 24] G. David Forney Jr and Pascal O. Vontobel. Par-tition functions of normal factor graphs. -
Human Perceptions of Fairness in Algorithmic Decision Making: A Case…
https://mlg.eng.cam.ac.uk/adrian/WWW18-HumanPerceptions.pdf19 Jun 2024: 47% 68% 70%B.S. 51% 32% 30%Liberal 57% 37% 33%Conservative 17% 24% 29%Moderate 21% 33% 34%Other 5% 6% 4%. ... Substance Abuse 4.84 0.08 0.07 0.10 0.24 0.07 0.68 0.26 0.22 0.20 0.07 0.284. -
Blind Justice: Fairness with Encrypted Sensitive Attributes
https://mlg.eng.cam.ac.uk/adrian/ICML18-BlindJustice.pdf19 Jun 2024: d features 51 62 7 24 23p sensitive attr. 1 1 7 1 1certification 802 ms 827 ms 288 ms 250 ms 765 mstraining 43 min 51 min 7 min 1 ... Bennett Capers, I. Blind justice. Yale Journal of Law &Humanities, 24:179, 2012. -
One-network Adversarial Fairness
https://mlg.eng.cam.ac.uk/adrian/AAAI2019_OneNetworkAdversarialFairness.pdf19 Jun 2024: log(1/δ). n(24). The term[4nI(x D0). 4nI(x D1). ]is what is estimated. ... From the latter note and (24), the two classifiers of theadversarial formulation proposed in (9) in the main docu-ment can be interpreted w.r.t. -
You Shouldn’t Trust Me: Learning Models WhichConceal Unfairness From…
https://mlg.eng.cam.ac.uk/adrian/ECAI20-You_Shouldn%E2%80%99t_Trust_Me.pdf19 Jun 2024: mostimportant feature) in 2D reduced input space (scikit-learn [24]’s PCA imple-mentation). ... 4765–4774, (2017). [24] F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. -
C:/Users/Adrian/Documents/GitHub/betheClean/docs/nb-UAI.dvi
https://mlg.eng.cam.ac.uk/adrian/nb-UAI.pdf19 Jun 2024: Approximating the Bethe Partition Function. Adrian WellerDepartment of Computer Science. Columbia UniversityNew York NY 10027. adrian@cs.columbia.edu. Tony JebaraDepartment of Computer Science. Columbia UniversityNew York NY 10027. jebara@cs.columbia -
The Geometry of Random Features Krzysztof Choromanski∗1 Mark…
https://mlg.eng.cam.ac.uk/adrian/geometry.pdf19 Jun 2024: The Geometry of Random Features. Krzysztof Choromanski1 Mark Rowland2 Tamas Sarlos1 Vikas Sindhwani1 Richard E. Turner2 Adrian Weller231Google Brain, NY 2University of Cambridge, UK 3The Alan Turing Institute, UK. Abstract. We present an in-depth -
Leader Stochastic Gradient Descent for DistributedTraining of Deep…
https://mlg.eng.cam.ac.uk/adrian/NeurIPS2019_LSGD_preprint.pdf19 Jun 2024: Landscape symmetries are commonin a plethora of non-convex problems [18, 19, 20, 21, 22], including deep learning [23, 24, 25, 26]. ... Understanding symmetries in deep networks. CoRR,abs/1511.01029, 2015. [24] A. Choromanska, M. -
Structured Evolution with Compact Architectures for Scalable Policy…
https://mlg.eng.cam.ac.uk/adrian/structured_icml_full.pdf19 Jun 2024: Structured Evolution with Compact Architecturesfor Scalable Policy Optimization. Krzysztof Choromanski 1 Mark Rowland 2 Vikas Sindhwani 1 Richard E. Turner 2 Adrian Weller 2 3. AbstractWe present a new method of blackbox optimiza-tion via gradient -
Tightness of LP Relaxations for Almost Balanced Models Adrian ...
https://mlg.eng.cam.ac.uk/adrian/tricam.pdf19 Jun 2024: Tightness of LP Relaxations for Almost Balanced Models. Adrian Weller Mark Rowland David SontagUniversity of Cambridge University of Cambridge New York University. Abstract. Linear programming (LP) relaxations are widelyused to attempt to identify a -
Clamping Variables and Approximate Inference Adrian WellerColumbia…
https://mlg.eng.cam.ac.uk/adrian/NeurIPS14-clamp.pdf19 Jun 2024: Journal of Automated Reasoning, 24(1-2):225–275, 2000. N. Ruozzi. The Bethe partition function of log-supermodular graphical models. -
From Parity to Preference-based Notionsof Fairness in Classification…
https://mlg.eng.cam.ac.uk/adrian/NeurIPS17-from-parity-to-preference.pdf19 Jun 2024: decisionoutcomes. A number of learning mechanisms have been proposed to achieve parity in treatment [24],. ... Angwin. https://github.com/propublica/compas-analysis, 2016. [24] B. T. Luong, S. Ruggieri, and F. -
A Unified Approach to Quantifying Algorithmic Unfairness: Measuring…
https://mlg.eng.cam.ac.uk/adrian/KDD2018_inequality_indices.pdf19 Jun 2024: 2.3 Axioms for Measuring InequalityBorrowing insights from the rich body of work on the axiomaticcharacterization of inequality indices in economics and social sci-ence [3, 10, 19, 24, 25, 28, -
Uprooting and Rerooting Higher-Order GraphicalModels Mark…
https://mlg.eng.cam.ac.uk/adrian/uprooting-higher-order.pdf19 Jun 2024: 4], which relates to generalized belief propagation,24) and MAP inference (using loopy belief propagation, LBP [9]). ... InArtificial Intelligence and Statistics (AISTATS), 2016. [24] J. Yedidia, W. Freeman, and Y. -
Unifying Orthogonal Monte Carlo Methods
https://mlg.eng.cam.ac.uk/adrian/ICML2019-unified.pdf19 Jun 2024: E[x̃λỹλx̃λvỹλv x̃. 2λỹ. 2λv. ] (24)We next show the following. Lemma A.7. -
Conditions Beyond Treewidth for Tightness of Higher-order LP…
https://mlg.eng.cam.ac.uk/adrian/conditions.pdf19 Jun 2024: Conditions Beyond Treewidth for Tightness of Higher-order LP Relaxations. Mark Rowland Aldo Pacchiano Adrian WellerUniversity of Cambridge UC Berkeley University of Cambridge. Abstract. Linear programming (LP) relaxations are a pop-ular method to -
Working Draft 1 Accountability of AI Under the Law: ...
https://mlg.eng.cam.ac.uk/adrian/SSRN-id3064761-Dec19.pdf19 Jun 2024: 24. Furthermore, an explanation must also provide the correct type of information in order for it to be useful. ... 24 Wachter, Right to Explanation, supra note 18. For a discussion about legibility of algorithmic systems more broadly, see Gianclaudio -
Geometrically Coupled Monte Carlo Sampling Mark Rowland∗University of …
https://mlg.eng.cam.ac.uk/adrian/NeurIPS18-gcmc.pdf19 Jun 2024: Geometrically Coupled Monte Carlo Sampling. Mark RowlandUniversity of Cambridgemr504@cam.ac.uk. Krzysztof ChoromanskiGoogle Brain Roboticskchoro@google.com. François ChalusUniversity of Cambridgechalusf3@gmail.com. Aldo PacchianoUniversity of -
Clamping Improves TRW and Mean Field Approximations Adrian Weller∗ ...
https://mlg.eng.cam.ac.uk/adrian/clamp_aistats_final.pdf19 Jun 2024: Journal of Automated Reasoning, 24(1-2):225–275, 2000. N. Ruozzi. The Bethe partition function of log-supermodular graphical models. -
The Unreasonable Effectiveness of StructuredRandom Orthogonal…
https://mlg.eng.cam.ac.uk/adrian/NeurIPS17-unreasonable-effectiveness.pdf19 Jun 2024: sin2(φ) cos2(ϕ1) cos2(ϕ2) sin2(ϕ1)cos2(ϕ2) sin. 2(ϕ1) sin2(φ) cos2(ϕ1). (24). This observation greatly simplifies the integral from Equation (22) involving the term -
Methods for Inference in Graphical Models
https://mlg.eng.cam.ac.uk/adrian/phd_FINAL.pdf19 Jun 2024: 3 Additional Background 24. 3.1 MAP Inference and Tractable Cases. 24.
Refine your results
clear all
Date
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.