A General Neural Causal Model for Interactive Recommendation

Survivor bias in observational data leads the optimization of recommender systems towards local optima. Currently most solutions re-mines existing human-system collaboration patterns to maximize longer-term satisfaction by reinforcement learning. However, from the causal perspective, mitigating survivor effects requires answering a counterfactual problem, which is generally unidentifiable and inestimable. In this work, we propose a neural causal model to achieve counterfactual inference. Specifically, we first build a learnable structural causal model based on its available graphical representations which qualitatively characterizes the preference transitions. Mitigation of the survivor bias is achieved though counterfactual consistency. To identify the consistency, we use the Gumbel-max function as structural constrains. To estimate the consistency, we apply reinforcement optimizations, and use Gumbel-Softmax as a trade-off to get a differentiable function. Both theoretical and empirical studies demonstrate the effectiveness of our solution.

Integrating Visual Foundation Models for Enhanced Robot Manipulation and Motion Planning: A Layered Approach

This paper presents a novel layered framework that integrates visual foundation models to improve robot manipulation tasks and motion planning. The framework consists of five layers: Perception, Cognition, Planning, Execution, and Learning. Using visual foundation models, we enhance the robot's perception of its environment, enabling more efficient task understanding and accurate motion planning. This approach allows for real-time adjustments and continual learning, leading to significant improvements in task execution. Experimental results demonstrate the effectiveness of the proposed framework in various robot manipulation tasks and motion planning scenarios, highlighting its potential for practical deployment in dynamic environments.

FocalDreamer: Text-driven 3D Editing via Focal-fusion Assembly

While text-3D editing has made significant strides in leveraging score distillation sampling, emerging approaches still fall short in delivering separable, precise and consistent outcomes that are vital to content creation. In response, we introduce FocalDreamer, a framework that merges base shape with editable parts according to text prompts for fine-grained editing within desired regions. Specifically, equipped with geometry union and dual-path rendering, FocalDreamer assembles independent 3D parts into a complete object, tailored for convenient instance reuse and part-wise control. We propose geometric focal loss and style consistency regularization, which encourage focal fusion and congruent overall appearance. Furthermore, FocalDreamer generates high-fidelity geometry and PBR textures which are compatible with widely-used graphics engines. Extensive experiments have highlighted the superior editing capabilities of FocalDreamer in both quantitative and qualitative evaluations.

Neuromorphic Hebbian learning with magnetic tunnel junction synapses

Neuromorphic computing aims to mimic both the function and structure of biological neural networks to provide artificial intelligence with extreme efficiency. Conventional approaches store synaptic weights in non-volatile memory devices with analog resistance states, permitting in-memory computation of neural network operations while avoiding the costs associated with transferring synaptic weights from a memory array. However, the use of analog resistance states for storing weights in neuromorphic systems is impeded by stochastic writing, weights drifting over time through stochastic processes, and limited endurance that reduces the precision of synapse weights. Here we propose and experimentally demonstrate neuromorphic networks that provide high-accuracy inference thanks to the binary resistance states of magnetic tunnel junctions (MTJs), while leveraging the analog nature of their stochastic spin-transfer torque (STT) switching for unsupervised Hebbian learning. We performed the first experimental demonstration of a neuromorphic network directly implemented with MTJ synapses, for both inference and spike-timing-dependent plasticity learning. We also demonstrated through simulation that the proposed system for unsupervised Hebbian learning with stochastic STT-MTJ synapses can achieve competitive accuracies for MNIST handwritten digit recognition. By appropriately applying neuromorphic principles through hardware-aware design, the proposed STT-MTJ neuromorphic learning networks provide a pathway toward artificial intelligence hardware that learns autonomously with extreme efficiency.

Fair Causal Feature Selection

Causal feature selection has recently received increasing attention in machine learning. Existing causal feature selection algorithms select unique causal features of a class variable as the optimal feature subset. However, a class variable usually has multiple states, and it is unfair to select the same causal features for different states of a class variable. To address this problem, we employ the class-specific mutual information to evaluate the causal information carried by each state of the class attribute, and theoretically analyze the unique relationship between each state and the causal features. Based on this, a Fair Causal Feature Selection algorithm (FairCFS) is proposed to fairly identifies the causal features for each state of the class variable. Specifically, FairCFS uses the pairwise comparisons of class-specific mutual information and the size of class-specific mutual information values from the perspective of each state, and follows a divide-and-conquer framework to find causal features. The correctness and application condition of FairCFS are theoretically proved, and extensive experiments are conducted to demonstrate the efficiency and superiority of FairCFS compared to the state-of-the-art approaches.

RWKV: Reinventing RNNs for the Transformer Era

Transformers have revolutionized almost all natural language processing (NLP) tasks but suffer from memory and computational complexity that scales quadratically with sequence length. In contrast, recurrent neural networks (RNNs) exhibit linear scaling in memory and computational requirements but struggle to match the same performance as Transformers due to limitations in parallelization and scalability. We propose a novel model architecture, Receptance Weighted Key Value (RWKV), that combines the efficient parallelizable training of Transformers with the efficient inference of RNNs. Our approach leverages a linear attention mechanism and allows us to formulate the model as either a Transformer or an RNN, which parallelizes computations during training and maintains constant computational and memory complexity during inference, leading to the first non-transformer architecture to be scaled to tens of billions of parameters. Our experiments reveal that RWKV performs on par with similarly sized Transformers, suggesting that future work can leverage this architecture to create more efficient models. This work presents a significant step towards reconciling the trade-offs between computational efficiency and model performance in sequence processing tasks.

Cascaded Logic Gates Based on High-Performance Ambipolar Dual-Gate WSe2 Thin Film Transistors

Ambipolar dual-gate transistors based on two-dimensional (2D) materials, such as graphene, carbon nanotubes, black phosphorus, and certain transition metal dichalcogenides (TMDs), enable reconfigurable logic circuits with suppressed off-state current. These circuits achieve the same logical output as CMOS with fewer transistors and offer greater flexibility in design. The primary challenge lies in the cascadability and power consumption of these logic gates with static CMOS-like connections. In this article, high-performance ambipolar dual-gate transistors based on tungsten diselenide (WSe2) are fabricated. A high on-off ratio of 10^8 and 10^6, a low off-state current of 100 to 300 fA, a negligible hysteresis, and an ideal subthreshold swing of 62 and 63 mV/dec are measured in the p- and n-type transport, respectively. For the first time, we demonstrate cascadable and cascaded logic gates using ambipolar TMD transistors with minimal static power consumption, including inverters, XOR, NAND, NOR, and buffers made by cascaded inverters. A thorough study of both the control gate and polarity gate behavior is conducted, which has previously been lacking. The noise margin of the logic gates is measured and analyzed. The large noise margin enables the implementation of VT-drop circuits, a type of logic with reduced transistor number and simplified circuit design. Finally, the speed performance of the VT-drop and other circuits built by dual-gate devices are qualitatively analyzed. This work lays the foundation for future developments in the field of ambipolar dual-gate TMD transistors, showing their potential for low-power, high-speed and more flexible logic circuits.

Resolving Task Confusion in Dynamic Expansion Architectures for Class Incremental Learning

The dynamic expansion architecture is becoming popular in class incremental learning, mainly due to its advantages in alleviating catastrophic forgetting. However, task confusion is not well assessed within this framework, e.g., the discrepancy between classes of different tasks is not well learned (i.e., inter-task confusion, ITC), and certain priority is still given to the latest class batch (i.e., old-new confusion, ONC). We empirically validate the side effects of the two types of confusion. Meanwhile, a novel solution called Task Correlated Incremental Learning (TCIL) is proposed to encourage discriminative and fair feature utilization across tasks. TCIL performs a multi-level knowledge distillation to propagate knowledge learned from old tasks to the new one. It establishes information flow paths at both feature and logit levels, enabling the learning to be aware of old classes. Besides, attention mechanism and classifier re-scoring are applied to generate more fair classification scores. We conduct extensive experiments on CIFAR100 and ImageNet100 datasets. The results demonstrate that TCIL consistently achieves state-of-the-art accuracy. It mitigates both ITC and ONC, while showing advantages in battle with catastrophic forgetting even no rehearsal memory is reserved.

A Dual-Arm Collaborative Framework for Dexterous Manipulation in Unstructured Environments with Contrastive Planning

Most object manipulation strategies for robots are based on the assumption that the object is rigid (i.e., with fixed geometry) and the goal's details have been fully specified (e.g., the exact target pose). However, there are many tasks that involve spatial relations in human environments where these conditions may be hard to satisfy, e.g., bending and placing a cable inside an unknown container. To develop advanced robotic manipulation capabilities in unstructured environments that avoid these assumptions, we propose a novel long-horizon framework that exploits contrastive planning in finding promising collaborative actions. Using simulation data collected by random actions, we learn an embedding model in a contrastive manner that encodes the spatio-temporal information from successful experiences, which facilitates the subgoal planning through clustering in the latent space. Based on the keypoint correspondence-based action parameterization, we design a leader-follower control scheme for the collaboration between dual arms. All models of our policy are automatically trained in simulation and can be directly transferred to real-world environments. To validate the proposed framework, we conduct a detailed experimental study on a complex scenario subject to environmental and reachability constraints in both simulation and real environments.

Gradient-based Neuromorphic Learning on Dynamical RRAM Arrays

We present MEMprop, the adoption of gradient-based learning to train fully memristive spiking neural networks (MSNNs). Our approach harnesses intrinsic device dynamics to trigger naturally arising voltage spikes. These spikes emitted by memristive dynamics are analog in nature, and thus fully differentiable, which eliminates the need for surrogate gradient methods that are prevalent in the spiking neural network (SNN) literature. Memristive neural networks typically either integrate memristors as synapses that map offline-trained networks, or otherwise rely on associative learning mechanisms to train networks of memristive neurons. We instead apply the backpropagation through time (BPTT) training algorithm directly on analog SPICE models of memristive neurons and synapses. Our implementation is fully memristive, in that synaptic weights and spiking neurons are both integrated on resistive RAM (RRAM) arrays without the need for additional circuits to implement spiking dynamics, e.g., analog-to-digital converters (ADCs) or thresholded comparators. As a result, higher-order electrophysical effects are fully exploited to use the state-driven dynamics of memristive neurons at run time. By moving towards non-approximate gradient-based learning, we obtain highly competitive accuracy amongst previously reported lightweight dense fully MSNNs on several benchmarks.