AI and science frontiers · 2026-10-22

DynaVelo puts RNA expression and transcription-factor motif accessibility from single-cell multiome data into one neural differential equation, learning a velocity field that can be perturbed rather than only drawn. In wild-type mouse germinal center B cells it recovers dark-zone cycling, the switch to the light zone and recycling; in Arid1a-heterozygous cells it nominated Ctcf, Stat3 and Bcl6 as rescue targets, and all three double-heterozygous mouse strains restored the dark-zone phenotype.

DynaVelo puts RNA expression and transcription-factor motif accessibility from single-cell multiome data into one neural differential equation, learning a velocity field that can be perturbed rather than only drawn. In wild-type mouse germinal center B cells it recovers dark-zone cycling, the switch to the light zone and recycling; in Arid1a-heterozygous cells it nominated Ctcf, Stat3 and Bcl6 as rescue targets, and all three double-heterozygous mouse strains restored the dark-zone phenotype.

TCR–pMHC binding prediction no longer has to start from a docked complex. Structure-based predictors co-fold the pair and read binding off the interface, so placement error changes which residues and contacts reach the model. EMBER folds each molecule as an unbound monomer, encodes both as rigid-motion-invariant residue graphs, and compares them through bidirectional attention over CDR3 and peptide residues. Across four folding tools and five epitope-disjoint folds it leads every metric, with 26% to 85% higher cross-tool ranking agreement than structure-based baselines.

TCR–pMHC binding prediction no longer has to start from a docked complex. Structure-based predictors co-fold the pair and read binding off the interface, so placement error changes which residues and contacts reach the model. EMBER folds each molecule as an unbound monomer, encodes both as rigid-motion-invariant residue graphs, and compares them through bidirectional attention over CDR3 and peptide residues. Across four folding tools and five epitope-disjoint folds it leads every metric, with 26% to 85% higher cross-tool ranking agreement than structure-based baselines.

Limited-angle atomic electron tomography leaves a missing wedge because specimens cannot be tilted far enough, degrading depth resolution and atomic localization. RESIRE-Net uses only the input dataset: it rotates the reconstructed volume, projects it at the experimental tilt angles, reconstructs each synthetic tilt series with RESIRE, and trains a 3D U-Net on the resulting pairs to approximate the inverse of the missing-wedge operator. On multislice-simulated metallic-glass and medium-entropy-alloy nanoparticles, 3D positional RMSD falls from 21 to 14 picometers and from 24 to 15 picometers, with atom recovery and elemental classification improving as well.

Limited-angle atomic electron tomography leaves a missing wedge because specimens cannot be tilted far enough, degrading depth resolution and atomic localization. RESIRE-Net uses only the input dataset: it rotates the reconstructed volume, projects it at the experimental tilt angles, reconstructs each synthetic tilt series with RESIRE, and trains a 3D U-Net on the resulting pairs to approximate the inverse of the missing-wedge operator. On multislice-simulated metallic-glass and medium-entropy-alloy nanoparticles, 3D positional RMSD falls from 21 to 14 picometers and from 24 to 15 picometers, with atom recovery and elemental classification improving as well.

Anthropic had its internal research model, Claude, run a largely autonomous 64-hour campaign designing minibinders against scorpion neurotoxins, generating 318,484 designs and selecting 180 for testing; with 31 more added by the team, Adaptyv Bio and Twist Bioscience tested all 211, and 17 met the binding criteria at both labs. The binders were shown to bind recombinant toxins only, with no neutralization or animal tests, so they are not an antivenom. All 211 designs and their data are released.

Anthropic had its internal research model, Claude, run a largely autonomous 64-hour campaign designing minibinders against scorpion neurotoxins, generating 318,484 designs and selecting 180 for testing; with 31 more added by the team, Adaptyv Bio and Twist Bioscience tested all 211, and 17 met the binding criteria at both labs. The binders were shown to bind recombinant toxins only, with no neutralization or animal tests, so they are not an antivenom. All 211 designs and their data are released.

OpenAI's alignment researchers add two monitorability metrics beyond g-mean²: one compares models only on task instances both support, reducing ranking bias from different instance mixtures, and one simulates a monitor as a rejection gate with retries, reporting bad-output probability, deployment probability, and the bad-output share among deployed outputs. In synthetic experiments, a known monitorability ordering is recovered by pair matching as capability spread grows.

OpenAI's alignment researchers add two monitorability metrics beyond g-mean²: one compares models only on task instances both support, reducing ranking bias from different instance mixtures, and one simulates a monitor as a rejection gate with retries, reporting bad-output probability, deployment probability, and the bad-output share among deployed outputs. In synthetic experiments, a known monitorability ordering is recovered by pair matching as capability spread grows.

In a stylized euro-area bank model of deposit pricing and bond reinvestment, large language model agents play the bank and four customer types, and equation-based decision rules are estimated from their choices. On simulations not used for estimation, the final rule keeps annual ROE within 0.31 percentage points in ten Claude simulated bank-years, meeting every fit criterion in seven; the three failures all involve deposit runs. With GPT-5.5 the gap stays within 0.71 points, but only six of eighteen simulated bank-years meet all criteria.

In a stylized euro-area bank model of deposit pricing and bond reinvestment, large language model agents play the bank and four customer types, and equation-based decision rules are estimated from their choices. On simulations not used for estimation, the final rule keeps annual ROE within 0.31 percentage points in ten Claude simulated bank-years, meeting every fit criterion in seven; the three failures all involve deposit runs. With GPT-5.5 the gap stays within 0.71 points, but only six of eighteen simulated bank-years meet all criteria.

The German Aerospace Center team applied physics-informed neural networks (PINNs) to reconstruct Eulerian flow fields from sparse 3D Lagrangian particle tracking data, comparing them systematically with polynomial binning and FlowFit on identical preprocessed data. Across homogeneous isotropic turbulence, a synthetic turbulent boundary layer, and wind-tunnel boundary-layer measurements, PINNs gave the lowest velocity and pressure errors over the investigated sparsity levels and degraded more slowly as particle density fell. FlowFit achieved lower vorticity errors at several densely seeded conditions and kept its divergence residual near machine precision.

The German Aerospace Center team applied physics-informed neural networks (PINNs) to reconstruct Eulerian flow fields from sparse 3D Lagrangian particle tracking data, comparing them systematically with polynomial binning and FlowFit on identical preprocessed data. Across homogeneous isotropic turbulence, a synthetic turbulent boundary layer, and wind-tunnel boundary-layer measurements, PINNs gave the lowest velocity and pressure errors over the investigated sparsity levels and degraded more slowly as particle density fell. FlowFit achieved lower vorticity errors at several densely seeded conditions and kept its divergence residual near machine precision.

Inferact and Google released vLLM TorchTPU, a backend that serves model definitions already in vLLM on Google Cloud TPUs without changing model code. Previously, bringing a model to TPU meant writing separate model code; this backend keeps vLLM's command line, OpenAI-compatible server, scheduler and KV cache manager, with TPU-specific work sitting underneath them. It compiles through torch.compile's TPU backend to XLA and uses custom JAX/Pallas kernels for attention, mixture-of-experts and quantized matmul.

Inferact and Google released vLLM TorchTPU, a backend that serves model definitions already in vLLM on Google Cloud TPUs without changing model code. Previously, bringing a model to TPU meant writing separate model code; this backend keeps vLLM's command line, OpenAI-compatible server, scheduler and KV cache manager, with TPU-specific work sitting underneath them. It compiles through torch.compile's TPU backend to XLA and uses custom JAX/Pallas kernels for attention, mixture-of-experts and quantized matmul.

Xiaomi released Xiaomi-Lens-0, a 4B multimodal model that decodes region masks from its own visual representations instead of relying on an extra segmentation encoder. Earlier multimodal models either took spatial detail from a separate segmentation encoder or kept the vision encoder frozen, so mask loss never reached the shared features. Xiaomi-Lens-0 pairs vision-encoder patch features with visual latents, the language model's states at generated mask tokens, in a lightweight decoder, so pixel-level supervision reaches the backbone shared with language generation; it gains 5.6 points on average over Qwen3.5 across all eight RefCOCO, RefCOCO+, and RefCOCOg splits and reaches the highest segmentation cIoU among compared models.

Xiaomi released Xiaomi-Lens-0, a 4B multimodal model that decodes region masks from its own visual representations instead of relying on an extra segmentation encoder. Earlier multimodal models either took spatial detail from a separate segmentation encoder or kept the vision encoder frozen, so mask loss never reached the shared features. Xiaomi-Lens-0 pairs vision-encoder patch features with visual latents, the language model's states at generated mask tokens, in a lightweight decoder, so pixel-level supervision reaches the backbone shared with language generation; it gains 5.6 points on average over Qwen3.5 across all eight RefCOCO, RefCOCO+, and RefCOCOg splits and reaches the highest segmentation cIoU among compared models.

The long-range memory bottleneck in gated recurrent networks is traced to the gates themselves having to be learned by the same recurrent computation, and adding short-dependency samples to training brings GRU and LSTM long-dependency test accuracy from near chance back to roughly ninety percent or higher, matching or exceeding Mamba and xLSTM.

The long-range memory bottleneck in gated recurrent networks is traced to the gates themselves having to be learned by the same recurrent computation, and adding short-dependency samples to training brings GRU and LSTM long-dependency test accuracy from near chance back to roughly ninety percent or higher, matching or exceeding Mamba and xLSTM.

Under a fixed prompt budget, matching a query to object category names places the target in context more often than pairwise geometry scoring, lifting R@10 on ScanRefer from 0.626 to 0.911, though geometry overtakes it when the query omits the target category word.

Under a fixed prompt budget, matching a query to object category names places the target in context more often than pairwise geometry scoring, lifting R@10 on ScanRefer from 0.626 to 0.911, though geometry overtakes it when the query omits the target category word.

Oilfield sites split each local-traffic update into secret shares so two non-colluding servers learn only accept or reject, holding macro-F1 at 0.641 against a 30% adaptive attacker where FedAvg falls to 0.046.

Oilfield sites split each local-traffic update into secret shares so two non-colluding servers learn only accept or reject, holding macro-F1 at 0.641 against a 30% adaptive attacker where FedAvg falls to 0.046.

Omnimodal large language models answer wrongly when audio and video interfere with each other, yet the question-relevant unimodal branch often still holds the right answer; a training-free inference-time intervention steers the fused representation back toward that branch on conflicting samples, raising CMM overall accuracy from 70.5% to 85.9%.

Omnimodal large language models answer wrongly when audio and video interfere with each other, yet the question-relevant unimodal branch often still holds the right answer; a training-free inference-time intervention steers the fused representation back toward that branch on conflicting samples, raising CMM overall accuracy from 70.5% to 85.9%.

Approximation error in pretrained attention concentrates in a few high-contribution query–key dimensions, and sharing quadratic-kernel coefficients by contribution group lowers frozen-attention relative MSE to 0.432 on Qwen2.5-0.5B and WikiText-103 perplexity from 31.71 to 27.50 after replacing middle-layer attention.

Approximation error in pretrained attention concentrates in a few high-contribution query–key dimensions, and sharing quadratic-kernel coefficients by contribution group lowers frozen-attention relative MSE to 0.432 on Qwen2.5-0.5B and WikiText-103 perplexity from 31.71 to 27.50 after replacing middle-layer attention.

FDE-Bench turns 49 real customer projects into end-to-end delivery tasks where agents must elicit unstated requirements from a simulated customer and iterate, and across 18 agent configurations the highest mean delivery score is only 0.5510 against an FDE reference of 1.0.

FDE-Bench turns 49 real customer projects into end-to-end delivery tasks where agents must elicit unstated requirements from a simulated customer and iterate, and across 18 agent configurations the highest mean delivery score is only 0.5510 against an FDE reference of 1.0.

Mixture-of-experts weight precision carries two separate costs, one for storage and one for decode; DuoRate stores a single 4.27 bpw copy and picks a read prefix per workload from that workload's own prefill, reaching 76.16 average accuracy at a 2.9 bpw read rate on Qwen3-30B-A3B, close to 4-bit formats.

Mixture-of-experts weight precision carries two separate costs, one for storage and one for decode; DuoRate stores a single 4.27 bpw copy and picks a read prefix per workload from that workload's own prefill, reaching 76.16 average accuracy at a 2.9 bpw read rate on Qwen3-30B-A3B, close to 4-bit formats.

Agents often recognize that a task lacks a prerequisite yet keep acting; a new evaluation scores recognition and warranted stopping separately, and harness search without model-weight updates raises strict self-awareness from 20% to 60%.

Agents often recognize that a task lacks a prerequisite yet keep acting; a new evaluation scores recognition and warranted stopping separately, and harness search without model-weight updates raises strict self-awareness from 20% to 60%.

A cross-architecture comparison of CNNs, ViTs and Vision Mamba uses human eye-tracking heatmaps as an independent benchmark to quantify how closely model saliency matches human gaze, finding that Mamba reaches the highest localization scores on chest X-rays while its saliency intensity is nearly uncorrelated with human fixations.

A cross-architecture comparison of CNNs, ViTs and Vision Mamba uses human eye-tracking heatmaps as an independent benchmark to quantify how closely model saliency matches human gaze, finding that Mamba reaches the highest localization scores on chest X-rays while its saliency intensity is nearly uncorrelated with human fixations.

When sensor failures spread along graph edges and value propagation loses its footing, PEMix conditions the distribution of missing node features on global structural representations, raising TADPOLE macro-F1 curve area from 0.7469 to 0.8243.

When sensor failures spread along graph edges and value propagation loses its footing, PEMix conditions the distribution of missing node features on global structural representations, raising TADPOLE macro-F1 curve area from 0.7469 to 0.8243.

Decoupling guided sampling from the optimization anchor lets a video flow model keep the gains of guided exploration under CFG-free single-branch inference while preserving the original semantics of inference-time guidance.

Decoupling guided sampling from the optimization anchor lets a video flow model keep the gains of guided exploration under CFG-free single-branch inference while preserving the original semantics of inference-time guidance.

PyTorch 2.15 brings FP8 tensors, a native grouped matmul and a batch of dedicated Metal kernels to the Mac's MPS backend, making local LLM decoding and video generation faster and lighter on memory. Previously many MPS operators went through MPSGraph, MoE decoding copied group offsets to the CPU and looped over groups, and tiled video VAE decoding pushed driver memory to 270 GiB. On an M3 Ultra, Qwen3.6-35B-A3B decoding rises from 9.4 to 61.3 tokens/s, and Wan2.2 VAE decoding peaks near 8 GiB.

PyTorch 2.15 brings FP8 tensors, a native grouped matmul and a batch of dedicated Metal kernels to the Mac's MPS backend, making local LLM decoding and video generation faster and lighter on memory. Previously many MPS operators went through MPSGraph, MoE decoding copied group offsets to the CPU and looped over groups, and tiled video VAE decoding pushed driver memory to 270 GiB. On an M3 Ultra, Qwen3.6-35B-A3B decoding rises from 9.4 to 61.3 tokens/s, and Wan2.2 VAE decoding peaks near 8 GiB.

Equilibrium Propagation (EqProp) updates each synapse by comparing a local quantity at the free and nudged equilibria, which normally requires extra memory or a duplicated network to hold the earlier value. Here that temporal comparison is carried by the synapse's own two-timescale dynamics: a volatile state relaxes toward the product of pre- and postsynaptic activities while a non-volatile state stores the learned parameter, and the volatile state's change is transferred to the non-volatile one through a global gate. On MNIST with a single-hidden-layer network, the phase-based and continual forms reach test errors of 2.01% and 2.24%, against 2.00% and 2.28% for standard EqProp, within one standard deviation.

Equilibrium Propagation (EqProp) updates each synapse by comparing a local quantity at the free and nudged equilibria, which normally requires extra memory or a duplicated network to hold the earlier value. Here that temporal comparison is carried by the synapse's own two-timescale dynamics: a volatile state relaxes toward the product of pre- and postsynaptic activities while a non-volatile state stores the learned parameter, and the volatile state's change is transferred to the non-volatile one through a global gate. On MNIST with a single-hidden-layer network, the phase-based and continual forms reach test errors of 2.01% and 2.24%, against 2.00% and 2.28% for standard EqProp, within one standard deviation.

Workflow selection for LLM agents is recast as a contextual bandit over a reusable pool, reaching an 81.37% average across five benchmarks while cutting MATH training-phase token overhead by at least 40.55%.

Workflow selection for LLM agents is recast as a contextual bandit over a reusable pool, reaching an 81.37% average across five benchmarks while cutting MATH training-phase token overhead by at least 40.55%.

On bidirectional Spanish–English speech-to-speech translation, PACT maps source pragmatic cues into the target-language expression space and injects them as continuous prompts into a frozen speech generator, giving the best overall performance among evaluated systems across content fidelity, voice preservation, and expression preservation.

On bidirectional Spanish–English speech-to-speech translation, PACT maps source pragmatic cues into the target-language expression space and injects them as continuous prompts into a frozen speech generator, giving the best overall performance among evaluated systems across content fidelity, voice preservation, and expression preservation.

Intermediate checkpoints of open-source models show that language abilities form in sequence: morphological structure begins before roughly 100M tokens, while semantic embeddings start only after 1B tokens and mature much later. The study evaluates Pythia and OLMo3 checkpoints under teacher forcing and probes divergence recognition and associative link generation with a modified divergent-association task and the Remote Associates Test. Larger training data helps divergence recognition, while co-occurrence in pre-training and repetition at inference weaken both abilities.

Intermediate checkpoints of open-source models show that language abilities form in sequence: morphological structure begins before roughly 100M tokens, while semantic embeddings start only after 1B tokens and mature much later. The study evaluates Pythia and OLMo3 checkpoints under teacher forcing and probes divergence recognition and associative link generation with a modified divergent-association task and the Remote Associates Test. Larger training data helps divergence recognition, while co-occurrence in pre-training and repetition at inference weaken both abilities.

Diffusion language models lose much of their accuracy when committing many tokens per step, and RePara reinforces short consecutive token blocks with group-relative policy optimization, raising LLaDA's GSM8K accuracy at Diff#16 from JustGRPO's 28.20 to 68.54.

Diffusion language models lose much of their accuracy when committing many tokens per step, and RePara reinforces short consecutive token blocks with group-relative policy optimization, raising LLaDA's GSM8K accuracy at Diff#16 from JustGRPO's 28.20 to 68.54.

A unified motion deblurring network learns blur as a convolution by predicting input-adaptive filters in Fourier space, raising GoPro PSNR from the strong EVSSM baseline's 33.26 dB to 34.02 dB for 8G more FLOPs.

A unified motion deblurring network learns blur as a convolution by predicting input-adaptive filters in Fourier space, raising GoPro PSNR from the strong EVSSM baseline's 33.26 dB to 34.02 dB for 8G more FLOPs.

Multi-agent routing policies can be cloned from execution logs, and a watermark that leaves model parameters and generated outputs untouched hides ownership signals in functionally equivalent rerouting decisions, remaining verifiable after up to 90% of logs are erased across four routing benchmarks.

Multi-agent routing policies can be cloned from execution logs, and a watermark that leaves model parameters and generated outputs untouched hides ownership signals in functionally equivalent rerouting decisions, remaining verifiable after up to 90% of logs are erased across four routing benchmarks.

LEST treats documented safety claims in model system cards as test specifications, finding that only 17 of 56 claims for Claude Opus 4.8 and GPT-5.5 are actually exercised by applicable benchmark prompts, and tracing validated counterexamples back to the claims they contradict.

LEST treats documented safety claims in model system cards as test specifications, finding that only 17 of 56 claims for Claude Opus 4.8 and GPT-5.5 are actually exercised by applicable benchmark prompts, and tracing validated counterexamples back to the claims they contradict.

Wavefunction evolution in a one-dimensional infinite square well is now predicted directly by a transformer, with probability normalization imposed as a hard architectural constraint on the output rather than a soft loss penalty. An unconstrained multilayer perceptron left total probability between 1.039418 and 1.045530, away from unity; the constrained model held the maximum probability deviation to 2.98×10⁻⁷ across ground, first-excited, Gaussian wavepacket and superposition initial states, with mean fidelity 0.999949. At matched spatial resolution, one forward pass takes roughly 0.0006 to 0.001 s, reaching a 22.35× speedup over sequential Crank–Nicolson at a 195-step horizon.

Wavefunction evolution in a one-dimensional infinite square well is now predicted directly by a transformer, with probability normalization imposed as a hard architectural constraint on the output rather than a soft loss penalty. An unconstrained multilayer perceptron left total probability between 1.039418 and 1.045530, away from unity; the constrained model held the maximum probability deviation to 2.98×10⁻⁷ across ground, first-excited, Gaussian wavepacket and superposition initial states, with mean fidelity 0.999949. At matched spatial resolution, one forward pass takes roughly 0.0006 to 0.001 s, reaching a 22.35× speedup over sequential Crank–Nicolson at a 195-step horizon.

Mastra's @mastra/mcp 2.0 now supports the MCP 2026-07-28 specification, which removes server sessions so every request is self-contained. Under the previous spec a client opened each connection with an initialize request, the server returned a session ID, and later requests had to carry it; a load balancer routing to another instance produced 400 session not found errors, worked around with sticky sessions or a shared store such as Redis. Now any instance can answer any request, a round-robin balancer suffices, and if an instance crashes the next request goes elsewhere while the client continues.

Mastra's @mastra/mcp 2.0 now supports the MCP 2026-07-28 specification, which removes server sessions so every request is self-contained. Under the previous spec a client opened each connection with an initialize request, the server returned a session ID, and later requests had to carry it; a load balancer routing to another instance produced 400 session not found errors, worked around with sticky sessions or a shared store such as Redis. Now any instance can answer any request, a round-robin balancer suffices, and if an instance crashes the next request goes elsewhere while the client continues.

Cloudflare's 27B multimodal decision model Clef is now available on Ollama, running locally or in Ollama's cloud. It takes a state plus a schema of typed questions and, in one pass over text, JSON or images, scores every answer to every question at once instead of generating a response word by word. On Cloudflare's published Jev Decision Index evaluations Clef beats Jev on most benchmarks, but trails it on When2Call and BRIGHT.

Cloudflare's 27B multimodal decision model Clef is now available on Ollama, running locally or in Ollama's cloud. It takes a state plus a schema of typed questions and, in one pass over text, JSON or images, scores every answer to every question at once instead of generating a response word by word. On Cloudflare's published Jev Decision Index evaluations Clef beats Jev on most benchmarks, but trails it on When2Call and BRIGHT.

Rotating sound source localization has long been troubled by ghost sources: time-domain ROSI and frequency-domain MCB both fail to remove Doppler-induced spurious sources when array geometry is unconstrained. A ResNet network takes an MCB-derived cross-spectral matrix as input and outputs the source intensity distribution directly, reaching up to 98.83% localization accuracy in simulations with 16 microphones, 500 to 2500 rpm and 4500 to 5000 Hz, with inference within 50 milliseconds per case.

Rotating sound source localization has long been troubled by ghost sources: time-domain ROSI and frequency-domain MCB both fail to remove Doppler-induced spurious sources when array geometry is unconstrained. A ResNet network takes an MCB-derived cross-spectral matrix as input and outputs the source intensity distribution directly, reaching up to 98.83% localization accuracy in simulations with 16 microphones, 500 to 2500 rpm and 4500 to 5000 Hz, with inference within 50 milliseconds per case.

Channel widths in a lightweight medical image classifier no longer have to be set by hand or found by architecture search. Fibonacci-Net allocated filters by a fixed second-order recurrence, leaving little control over capacity. Scaled m-BonacciNet derives a scaling coefficient from the dominant root of the m-Bonacci recurrence, narrows the original channel vector, and uses depthwise separable convolutions in the last blocks; on 15-class brain MRI, the Tribonacci variant reached 96.09% accuracy and 0.9581 macro F1 with about 0.85 million parameters and 365.69 million FLOPs, though some variants lost accuracy on 4-class brain MRI and CKD ultrasound.

Channel widths in a lightweight medical image classifier no longer have to be set by hand or found by architecture search. Fibonacci-Net allocated filters by a fixed second-order recurrence, leaving little control over capacity. Scaled m-BonacciNet derives a scaling coefficient from the dominant root of the m-Bonacci recurrence, narrows the original channel vector, and uses depthwise separable convolutions in the last blocks; on 15-class brain MRI, the Tribonacci variant reached 96.09% accuracy and 0.9581 macro F1 with about 0.85 million parameters and 365.69 million FLOPs, though some variants lost accuracy on 4-class brain MRI and CKD ultrasound.

IoT device authentication is shifting from a one-time check of fixed credentials to continuous checking of device behavior. Static credentials and centralized trust miss anomalies after a device is compromised and keep exposing identifiable attributes. A combined scheme uses hardware behavioral fingerprints, federated learning with differential privacy, a lightweight permissioned chain and agentic orchestration; on 11 Raspberry Pi devices in the PP-IDE dataset, differential privacy lowered membership inference attack success from 48.2% to 12.4%, with 99.2% authentication accuracy.

IoT device authentication is shifting from a one-time check of fixed credentials to continuous checking of device behavior. Static credentials and centralized trust miss anomalies after a device is compromised and keep exposing identifiable attributes. A combined scheme uses hardware behavioral fingerprints, federated learning with differential privacy, a lightweight permissioned chain and agentic orchestration; on 11 Raspberry Pi devices in the PP-IDE dataset, differential privacy lowered membership inference attack success from 48.2% to 12.4%, with 99.2% authentication accuracy.

AMD is extending FSR Upscaling 4.1 from Radeon RX 9000 series GPUs to RX 6000 series cards, Ryzen AI Max processors, Ryzen AI 9 laptops and several handhelds, with Xbox Series X|S coming soon. The technology reconstructs crisp visuals from lower-resolution frames using machine learning, and on RX 6000 it delivers a significant image-quality leap over the previously available FSR 3.1, without requiring new components.

AMD is extending FSR Upscaling 4.1 from Radeon RX 9000 series GPUs to RX 6000 series cards, Ryzen AI Max processors, Ryzen AI 9 laptops and several handhelds, with Xbox Series X|S coming soon. The technology reconstructs crisp visuals from lower-resolution frames using machine learning, and on RX 6000 it delivers a significant image-quality leap over the previously available FSR 3.1, without requiring new components.