Class (philosophy)

About: Class (philosophy) is a research topic. Over the lifetime, 821 publications have been published within this topic receiving 28000 citations.

Inductive inference in the limit

Abstract: There is an enormous and interesting theoretical literature1 on inductive inference which remains largely unknown to philosophers of science, even though a philosopher, Hilary Putnam, may be said to have initiated it2. The work in this tradition concerns algorithms for inferring recursive functions from finite samples of their graphs. Informally, an inductive machine is an algorithm which is given larger and larger samples of the graph of a partial or total recursive function which the machine attempts to identify. The machine or algorithm produces at various stages a program which computes a recursive function. The machine is said to identify the target function if at some point it produces a program which computes that very function, and thereafter, no matter how much more of the graph of the target function it sees, continues to produce the same program. A weaker notion, that of behaviorally correct identification, does not require that in order to identify the target function the machine converge to a single program. Instead it requires only that the machine converge to a (possibly infinite) set of programs, all of which compute the target function. A class of functions is said to be identified by a machine if the machine identifies every function in the class. This framework has been adapted to characterize abstract versions of learning languages in the limit. A natural question is whether a similar framework can be developed for learning first order theories. What follows is an attempt to present such a conception, and to describe what place Hempel’s ideas about confirmation have within it. There are many definitions, some examples, no theorems, and a great deal left to be investigated.

Dependency-based grouping to establish class identity

Abstract: A method for using dependency-based grouping to establish class identity comprises categorizing a plurality of classes into a set of class groups based at least in part on one or more dependencies between the classes, and generating metadata to be use for loading the classes, where the metadata includes a mapping between the set of class groups and the plurality of classes. The metadata may also include respective signatures for class groups and/or the individual classes. The method may also include validating, using at least a portion of the metadata, the identity of a particular version of a class of the plurality of classes, prior to loading the version for execution.

Decoupled Knowledge Distillation

Abstract: State-of-the-art distillation methods are mainly based on distilling deep features from intermediate layers, while the significance of logit distillation is greatly overlooked. To provide a novel viewpoint to study logit distillation, we re-formulate the classical KD loss into two parts, i.e., target class knowledge distillation (TCKD) and non-target class knowledge distillation (NCKD). We empirically investigate and prove the effects of the two parts: TCKD transfers knowledge concerning the “difficulty” of training samples, while NCKD is the prominent reason why logit distillation works. More importantly, we reveal that the classical KD loss is a coupled formulation, which (1) suppresses the effectiveness of NCKD and (2) limits the flexibility to balance these two parts. To address these issues, we present Decoupled Knowledge Distillation (DKD), enabling TCKD and NCKD to play their roles more efficiently and flexibly. Compared with complex feature-based methods, our DKD achieves comparable or even better results and has better training efficiency on CIFAR-100, ImageNet, and MS-COCO datasets for image classification and object detection tasks. This paper proves the great potential of logit distillation, and we hope it will be helpful for future research. The code is available at https://github.com/megviiresearch/mdistiller.

Few-Shot Learning for Dermatological Disease Diagnosis.

Abstract: We consider the problem of clinical image classification for the purpose of aiding doctors in dermatological disease diagnosis. Diagnosis of dermatological conditions from images poses two major challenges for standard off-the-shelf techniques: First, the distribution of real-world dermatological datasets is typically long-tailed. Second, intra-class variability is large. To address the first issue, we formulate the problem as low-shot learning, where once deployed, a base classifier must rapidly generalize to diagnose novel conditions given very few labeled examples. To model intra-class variability effectively, we propose Prototypical Clustering Networks (PCN), an extension to Prototypical Networks (Snell et al., 2017) that learns a mixture of “prototypes” for each class. Prototypes are initialized for each class via clustering and refined via an online update scheme. Classification is performed by measuring similarity to a weighted combination of prototypes within a class, where the weights are the inferred cluster responsibilities. We demonstrate the strengths of our approach in effective diagnosis on a realistic dataset of dermatological conditions.

A Dynamic Frame Selection Framework for Fast Video Recognition.

Abstract: We introduce AdaFrame, a conditional computation framework that adaptively selects relevant frames on a per-input basis for fast video recognition. AdaFrame, which contains a Long Short-Term Memory augmented with a global memory to provide context information, operates as an agent to interact with video sequences aiming to search over time which frames to use. Trained with policy search methods, at each time step, AdaFrame computes a prediction, decides where to observe next, and estimates a utility, i.e., expected future rewards, of viewing more frames in the future. Exploring predicted utilities at testing time, AdaFrame is able to achieve adaptive lookahead inference so as to minimize the overall computational cost without incurring a degradation in accuracy. We conduct extensive experiments on two large-scale video benchmarks, FCVID and ActivityNet. With a vanilla ResNet-101 model, AdaFrame achieves similar performance of using all frames while only requiring, on average, 8.21 and 8.65 frames on FCVID and ActivityNet, respectively. We also demonstrate AdaFrame is compatible with modern 2D and 3D networks for video recognition. Furthermore, we show, among other things, learned frame usage can reflect the difficulty of making prediction decisions both at instance-level within the same class and at class-level among different categories.